< prev index next >

src/hotspot/cpu/aarch64/c1_MacroAssembler_aarch64.cpp

Print this page

 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 "c1/c1_MacroAssembler.hpp"
 28 #include "c1/c1_Runtime1.hpp"
 29 #include "gc/shared/barrierSetAssembler.hpp"
 30 #include "gc/shared/collectedHeap.hpp"


 31 #include "gc/shared/tlab_globals.hpp"
 32 #include "interpreter/interpreter.hpp"
 33 #include "oops/arrayOop.hpp"
 34 #include "oops/markWord.hpp"
 35 #include "runtime/basicLock.hpp"
 36 #include "runtime/os.hpp"
 37 #include "runtime/sharedRuntime.hpp"
 38 #include "runtime/stubRoutines.hpp"
 39 
 40 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result,
 41                                   FloatRegister f0, FloatRegister f1,
 42                                   Register result)
 43 {
 44   Label done;
 45   if (is_float) {
 46     fcmps(f0, f1);
 47   } else {
 48     fcmpd(f0, f1);
 49   }
 50   if (unordered_result < 0) {

 71   // save object being locked into the BasicObjectLock
 72   str(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
 73 
 74   null_check_offset = offset();
 75 
 76   if (DiagnoseSyncOnValueBasedClasses != 0) {
 77     load_klass(hdr, obj);
 78     ldrb(hdr, Address(hdr, Klass::misc_flags_offset()));
 79     tst(hdr, KlassFlags::_misc_is_value_based_class);
 80     br(Assembler::NE, slow_case);
 81   }
 82 
 83   if (LockingMode == LM_LIGHTWEIGHT) {
 84     lightweight_lock(disp_hdr, obj, hdr, temp, rscratch2, slow_case);
 85   } else if (LockingMode == LM_LEGACY) {
 86     Label done;
 87     // Load object header
 88     ldr(hdr, Address(obj, hdr_offset));
 89     // and mark it as unlocked
 90     orr(hdr, hdr, markWord::unlocked_value);






 91     // save unlocked object header into the displaced header location on the stack
 92     str(hdr, Address(disp_hdr, 0));
 93     // test if object header is still the same (i.e. unlocked), and if so, store the
 94     // displaced header address in the object header - if it is not the same, get the
 95     // object header instead
 96     lea(rscratch2, Address(obj, hdr_offset));
 97     cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/nullptr);
 98     // if the object header was the same, we're done
 99     // if the object header was not the same, it is now in the hdr register
100     // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
101     //
102     // 1) (hdr & aligned_mask) == 0
103     // 2) sp <= hdr
104     // 3) hdr <= sp + page_size
105     //
106     // these 3 tests can be done by evaluating the following expression:
107     //
108     // (hdr - sp) & (aligned_mask - page_size)
109     //
110     // assuming both the stack pointer and page_size have their least

158       cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
159     }
160     // done
161     bind(done);
162   }
163   decrement(Address(rthread, JavaThread::held_monitor_count_offset()));
164 }
165 
166 
167 // Defines obj, preserves var_size_in_bytes
168 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
169   if (UseTLAB) {
170     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
171   } else {
172     b(slow_case);
173   }
174 }
175 
176 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
177   assert_different_registers(obj, klass, len);
178   // This assumes that all prototype bits fit in an int32_t
179   mov(t1, (int32_t)(intptr_t)markWord::prototype().value());






180   str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
181 
182   if (UseCompressedClassPointers) { // Take care not to kill klass
183     encode_klass_not_null(t1, klass);
184     strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
185   } else {
186     str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
187   }
188 
189   if (len->is_valid()) {
190     strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
191     int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
192     if (!is_aligned(base_offset, BytesPerWord)) {
193       assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
194       // Clear gap/first 4 bytes following the length field.
195       strw(zr, Address(obj, base_offset));
196     }
197   } else if (UseCompressedClassPointers) {
198     store_klass_gap(obj, zr);
199   }

297   // following the length field in initialize_header().
298   int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
299   // clear rest of allocated space
300   if (zero_array) {
301     initialize_body(obj, arr_size, base_offset, t1, t2);
302   }
303   if (Compilation::current()->bailed_out()) {
304     return;
305   }
306 
307   membar(StoreStore);
308 
309   if (CURRENT_ENV->dtrace_alloc_probes()) {
310     assert(obj == r0, "must be");
311     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
312   }
313 
314   verify_oop(obj);
315 }
316 
317 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes) {
318   assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");











319   // Make sure there is enough stack space for this method's activation.
320   // Note that we do this before creating a frame.

321   generate_stack_overflow_check(bang_size_in_bytes);
322   MacroAssembler::build_frame(framesize);

323 
324   // Insert nmethod entry barrier into frame.
325   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
326   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
327 }
328 
329 void C1_MacroAssembler::remove_frame(int framesize) {
330   MacroAssembler::remove_frame(framesize);


331 }
332 
333 
334 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
335   // If we have to make this method not-entrant we'll overwrite its
336   // first instruction with a jump.  For this action to be legal we
337   // must ensure that this first instruction is a B, BL, NOP, BKPT,
338   // SVC, HVC, or SMC.  Make it a NOP.
339   nop();






























































340 }
341 

342 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
343   // rfp, + 0: link
344   //     + 1: return address
345   //     + 2: argument with offset 0
346   //     + 3: argument with offset 1
347   //     + 4: ...
348 
349   ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
350 }
351 
352 #ifndef PRODUCT
353 
354 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
355   if (!VerifyOops) return;
356   verify_oop_addr(Address(sp, stack_offset));
357 }
358 
359 void C1_MacroAssembler::verify_not_null_oop(Register r) {
360   if (!VerifyOops) return;
361   Label not_null;

 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 "c1/c1_MacroAssembler.hpp"
 28 #include "c1/c1_Runtime1.hpp"
 29 #include "gc/shared/barrierSetAssembler.hpp"
 30 #include "gc/shared/collectedHeap.hpp"
 31 #include "gc/shared/barrierSet.hpp"
 32 #include "gc/shared/barrierSetAssembler.hpp"
 33 #include "gc/shared/tlab_globals.hpp"
 34 #include "interpreter/interpreter.hpp"
 35 #include "oops/arrayOop.hpp"
 36 #include "oops/markWord.hpp"
 37 #include "runtime/basicLock.hpp"
 38 #include "runtime/os.hpp"
 39 #include "runtime/sharedRuntime.hpp"
 40 #include "runtime/stubRoutines.hpp"
 41 
 42 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result,
 43                                   FloatRegister f0, FloatRegister f1,
 44                                   Register result)
 45 {
 46   Label done;
 47   if (is_float) {
 48     fcmps(f0, f1);
 49   } else {
 50     fcmpd(f0, f1);
 51   }
 52   if (unordered_result < 0) {

 73   // save object being locked into the BasicObjectLock
 74   str(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
 75 
 76   null_check_offset = offset();
 77 
 78   if (DiagnoseSyncOnValueBasedClasses != 0) {
 79     load_klass(hdr, obj);
 80     ldrb(hdr, Address(hdr, Klass::misc_flags_offset()));
 81     tst(hdr, KlassFlags::_misc_is_value_based_class);
 82     br(Assembler::NE, slow_case);
 83   }
 84 
 85   if (LockingMode == LM_LIGHTWEIGHT) {
 86     lightweight_lock(disp_hdr, obj, hdr, temp, rscratch2, slow_case);
 87   } else if (LockingMode == LM_LEGACY) {
 88     Label done;
 89     // Load object header
 90     ldr(hdr, Address(obj, hdr_offset));
 91     // and mark it as unlocked
 92     orr(hdr, hdr, markWord::unlocked_value);
 93 
 94     if (EnableValhalla) {
 95       // Mask always_locked bit such that we go to the slow path if object is an inline type
 96       andr(hdr, hdr, ~markWord::inline_type_bit_in_place);
 97     }
 98 
 99     // save unlocked object header into the displaced header location on the stack
100     str(hdr, Address(disp_hdr, 0));
101     // test if object header is still the same (i.e. unlocked), and if so, store the
102     // displaced header address in the object header - if it is not the same, get the
103     // object header instead
104     lea(rscratch2, Address(obj, hdr_offset));
105     cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/nullptr);
106     // if the object header was the same, we're done
107     // if the object header was not the same, it is now in the hdr register
108     // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
109     //
110     // 1) (hdr & aligned_mask) == 0
111     // 2) sp <= hdr
112     // 3) hdr <= sp + page_size
113     //
114     // these 3 tests can be done by evaluating the following expression:
115     //
116     // (hdr - sp) & (aligned_mask - page_size)
117     //
118     // assuming both the stack pointer and page_size have their least

166       cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
167     }
168     // done
169     bind(done);
170   }
171   decrement(Address(rthread, JavaThread::held_monitor_count_offset()));
172 }
173 
174 
175 // Defines obj, preserves var_size_in_bytes
176 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
177   if (UseTLAB) {
178     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
179   } else {
180     b(slow_case);
181   }
182 }
183 
184 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
185   assert_different_registers(obj, klass, len);
186   if (EnableValhalla) {
187     // Need to copy markWord::prototype header for klass
188     assert_different_registers(obj, klass, len, t1, t2);
189     ldr(t1, Address(klass, Klass::prototype_header_offset()));
190   } else {
191     // This assumes that all prototype bits fit in an int32_t
192     mov(t1, (int32_t)(intptr_t)markWord::prototype().value());
193   }
194   str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
195 
196   if (UseCompressedClassPointers) { // Take care not to kill klass
197     encode_klass_not_null(t1, klass);
198     strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
199   } else {
200     str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
201   }
202 
203   if (len->is_valid()) {
204     strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
205     int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
206     if (!is_aligned(base_offset, BytesPerWord)) {
207       assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
208       // Clear gap/first 4 bytes following the length field.
209       strw(zr, Address(obj, base_offset));
210     }
211   } else if (UseCompressedClassPointers) {
212     store_klass_gap(obj, zr);
213   }

311   // following the length field in initialize_header().
312   int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
313   // clear rest of allocated space
314   if (zero_array) {
315     initialize_body(obj, arr_size, base_offset, t1, t2);
316   }
317   if (Compilation::current()->bailed_out()) {
318     return;
319   }
320 
321   membar(StoreStore);
322 
323   if (CURRENT_ENV->dtrace_alloc_probes()) {
324     assert(obj == r0, "must be");
325     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
326   }
327 
328   verify_oop(obj);
329 }
330 
331 void C1_MacroAssembler::build_frame_helper(int frame_size_in_bytes, int sp_offset_for_orig_pc, int sp_inc, bool reset_orig_pc, bool needs_stack_repair) {
332   MacroAssembler::build_frame(frame_size_in_bytes);
333 
334   if (needs_stack_repair) {
335     save_stack_increment(sp_inc, frame_size_in_bytes);
336   }
337   if (reset_orig_pc) {
338     // Zero orig_pc to detect deoptimization during buffering in the entry points
339     str(zr, Address(sp, sp_offset_for_orig_pc));
340   }
341 }
342 
343 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes, int sp_offset_for_orig_pc, bool needs_stack_repair, bool has_scalarized_args, Label* verified_inline_entry_label) {
344   // Make sure there is enough stack space for this method's activation.
345   // Note that we do this before creating a frame.
346   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
347   generate_stack_overflow_check(bang_size_in_bytes);
348 
349   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, has_scalarized_args, needs_stack_repair);
350 
351   // Insert nmethod entry barrier into frame.
352   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
353   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);

354 
355   if (verified_inline_entry_label != nullptr) {
356     // Jump here from the scalarized entry points that already created the frame.
357     bind(*verified_inline_entry_label);
358   }
359 }
360 

361 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
362   // If we have to make this method not-entrant we'll overwrite its
363   // first instruction with a jump.  For this action to be legal we
364   // must ensure that this first instruction is a B, BL, NOP, BKPT,
365   // SVC, HVC, or SMC.  Make it a NOP.
366   nop();
367   if (C1Breakpoint) brk(1);
368 }
369 
370 int C1_MacroAssembler::scalarized_entry(const CompiledEntrySignature* ces, int frame_size_in_bytes, int bang_size_in_bytes, int sp_offset_for_orig_pc, Label& verified_inline_entry_label, bool is_inline_ro_entry) {
371   assert(InlineTypePassFieldsAsArgs, "sanity");
372   // Make sure there is enough stack space for this method's activation.
373   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
374   generate_stack_overflow_check(bang_size_in_bytes);
375 
376   GrowableArray<SigEntry>* sig    = ces->sig();
377   GrowableArray<SigEntry>* sig_cc = is_inline_ro_entry ? ces->sig_cc_ro() : ces->sig_cc();
378   VMRegPair* regs      = ces->regs();
379   VMRegPair* regs_cc   = is_inline_ro_entry ? ces->regs_cc_ro() : ces->regs_cc();
380   int args_on_stack    = ces->args_on_stack();
381   int args_on_stack_cc = is_inline_ro_entry ? ces->args_on_stack_cc_ro() : ces->args_on_stack_cc();
382 
383   assert(sig->length() <= sig_cc->length(), "Zero-sized inline class not allowed!");
384   BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sig_cc->length());
385   int args_passed = sig->length();
386   int args_passed_cc = SigEntry::fill_sig_bt(sig_cc, sig_bt);
387 
388   // Create a temp frame so we can call into the runtime. It must be properly set up to accommodate GC.
389   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, true, ces->c1_needs_stack_repair());
390 
391   // The runtime call might safepoint, make sure nmethod entry barrier is executed
392   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
393   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
394   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
395 
396   // FIXME -- call runtime only if we cannot in-line allocate all the incoming inline type args.
397   mov(r19, (intptr_t) ces->method());
398   if (is_inline_ro_entry) {
399     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::buffer_inline_args_no_receiver_id)));
400   } else {
401     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::buffer_inline_args_id)));
402   }
403   int rt_call_offset = offset();
404 
405   // The runtime call returns the new array in r20 instead of the usual r0
406   // because r0 is also j_rarg7 which may be holding a live argument here.
407   Register val_array = r20;
408 
409   // Remove the temp frame
410   MacroAssembler::remove_frame(frame_size_in_bytes);
411 
412   // Check if we need to extend the stack for packing
413   int sp_inc = 0;
414   if (args_on_stack > args_on_stack_cc) {
415     sp_inc = extend_stack_for_inline_args(args_on_stack);
416   }
417 
418   shuffle_inline_args(true, is_inline_ro_entry, sig_cc,
419                       args_passed_cc, args_on_stack_cc, regs_cc, // from
420                       args_passed, args_on_stack, regs,          // to
421                       sp_inc, val_array);
422 
423   // Create the real frame. Below jump will then skip over the stack banging and frame
424   // setup code in the verified_inline_entry (which has a different real_frame_size).
425   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, sp_inc, false, ces->c1_needs_stack_repair());
426 
427   b(verified_inline_entry_label);
428   return rt_call_offset;
429 }
430 
431 
432 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
433   // rfp, + 0: link
434   //     + 1: return address
435   //     + 2: argument with offset 0
436   //     + 3: argument with offset 1
437   //     + 4: ...
438 
439   ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
440 }
441 
442 #ifndef PRODUCT
443 
444 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
445   if (!VerifyOops) return;
446   verify_oop_addr(Address(sp, stack_offset));
447 }
448 
449 void C1_MacroAssembler::verify_not_null_oop(Register r) {
450   if (!VerifyOops) return;
451   Label not_null;
< prev index next >