rev 54670 : Port of valuetypes to aarch64

   1 /*
   2  * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, Red Hat Inc. 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 "c1/c1_MacroAssembler.hpp"
  28 #include "c1/c1_Runtime1.hpp"
  29 #include "classfile/systemDictionary.hpp"
  30 #include "gc/shared/collectedHeap.hpp"
  31 #include "interpreter/interpreter.hpp"
  32 #include "oops/arrayOop.hpp"
  33 #include "oops/markOop.hpp"
  34 #include "runtime/basicLock.hpp"
  35 #include "runtime/biasedLocking.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) {
  51     // we want -1 for unordered or less than, 0 for equal and 1 for
  52     // greater than.
  53     cset(result, NE);  // Not equal or unordered
  54     cneg(result, result, LT);  // Less than or unordered
  55   } else {
  56     // we want -1 for less than, 0 for equal and 1 for unordered or
  57     // greater than.
  58     cset(result, NE);  // Not equal or unordered
  59     cneg(result, result, LO);  // Less than
  60   }
  61 }
  62 
  63 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
  64   const int aligned_mask = BytesPerWord -1;
  65   const int hdr_offset = oopDesc::mark_offset_in_bytes();
  66   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
  67   Label done;
  68   int null_check_offset = -1;
  69 
  70   verify_oop(obj);
  71 
  72   // save object being locked into the BasicObjectLock
  73   str(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
  74 
  75   if (UseBiasedLocking) {
  76     assert(scratch != noreg, "should have scratch register at this point");
  77     null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case);
  78   } else {
  79     null_check_offset = offset();
  80   }
  81 
  82   // Load object header
  83   ldr(hdr, Address(obj, hdr_offset));
  84   // and mark it as unlocked
  85   orr(hdr, hdr, markOopDesc::unlocked_value);
  86 
  87   if (EnableValhalla && !UseBiasedLocking) {
  88     // Mask always_locked bit such that we go to the slow path if object is a value type
  89     andr(hdr, hdr, ~markOopDesc::biased_lock_bit_in_place);
  90   }
  91 
  92   // save unlocked object header into the displaced header location on the stack
  93   str(hdr, Address(disp_hdr, 0));
  94   // test if object header is still the same (i.e. unlocked), and if so, store the
  95   // displaced header address in the object header - if it is not the same, get the
  96   // object header instead
  97   lea(rscratch2, Address(obj, hdr_offset));
  98   cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/NULL);
  99   // if the object header was the same, we're done
 100   // if the object header was not the same, it is now in the hdr register
 101   // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
 102   //
 103   // 1) (hdr & aligned_mask) == 0
 104   // 2) sp <= hdr
 105   // 3) hdr <= sp + page_size
 106   //
 107   // these 3 tests can be done by evaluating the following expression:
 108   //
 109   // (hdr - sp) & (aligned_mask - page_size)
 110   //
 111   // assuming both the stack pointer and page_size have their least
 112   // significant 2 bits cleared and page_size is a power of 2
 113   mov(rscratch1, sp);
 114   sub(hdr, hdr, rscratch1);
 115   ands(hdr, hdr, aligned_mask - os::vm_page_size());
 116   // for recursive locking, the result is zero => save it in the displaced header
 117   // location (NULL in the displaced hdr location indicates recursive locking)
 118   str(hdr, Address(disp_hdr, 0));
 119   // otherwise we don't care about the result and handle locking via runtime call
 120   cbnz(hdr, slow_case);
 121   // done
 122   bind(done);
 123   if (PrintBiasedLockingStatistics) {
 124     lea(rscratch2, ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
 125     addmw(Address(rscratch2, 0), 1, rscratch1);
 126   }
 127   return null_check_offset;
 128 }
 129 
 130 
 131 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
 132   const int aligned_mask = BytesPerWord -1;
 133   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 134   assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
 135   Label done;
 136 
 137   if (UseBiasedLocking) {
 138     // load object
 139     ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
 140     biased_locking_exit(obj, hdr, done);
 141   }
 142 
 143   // load displaced header
 144   ldr(hdr, Address(disp_hdr, 0));
 145   // if the loaded hdr is NULL we had recursive locking
 146   // if we had recursive locking, we are done
 147   cbz(hdr, done);
 148   if (!UseBiasedLocking) {
 149     // load object
 150     ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
 151   }
 152   verify_oop(obj);
 153   // test if object header is pointing to the displaced header, and if so, restore
 154   // the displaced header in the object - if the object header is not pointing to
 155   // the displaced header, get the object header instead
 156   // if the object header was not pointing to the displaced header,
 157   // we do unlocking via runtime call
 158   if (hdr_offset) {
 159     lea(rscratch1, Address(obj, hdr_offset));
 160     cmpxchgptr(disp_hdr, hdr, rscratch1, rscratch2, done, &slow_case);
 161   } else {
 162     cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
 163   }
 164   // done
 165   bind(done);
 166 }
 167 
 168 
 169 // Defines obj, preserves var_size_in_bytes
 170 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
 171   if (UseTLAB) {
 172     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 173   } else {
 174     eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
 175   }
 176 }
 177 
 178 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
 179   assert_different_registers(obj, klass, len);
 180   if (UseBiasedLocking && !len->is_valid()) {
 181     assert_different_registers(obj, klass, len, t1, t2);
 182     ldr(t1, Address(klass, Klass::prototype_header_offset()));
 183   } else {
 184     // This assumes that all prototype bits fit in an int32_t
 185     mov(t1, (int32_t)(intptr_t)markOopDesc::prototype());
 186   }
 187   str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
 188 
 189   if (UseCompressedClassPointers) { // Take care not to kill klass
 190     encode_klass_not_null(t1, klass);
 191     strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
 192   } else {
 193     str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 194   }
 195 
 196   if (len->is_valid()) {
 197     strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
 198   } else if (UseCompressedClassPointers) {
 199     store_klass_gap(obj, zr);
 200   }
 201 }
 202 
 203 // preserves obj, destroys len_in_bytes
 204 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
 205   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
 206   Label done;
 207 
 208   // len_in_bytes is positive and ptr sized
 209   subs(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
 210   br(Assembler::EQ, done);
 211 
 212   // Preserve obj
 213   if (hdr_size_in_bytes)
 214     add(obj, obj, hdr_size_in_bytes);
 215   zero_memory(obj, len_in_bytes, t1);
 216   if (hdr_size_in_bytes)
 217     sub(obj, obj, hdr_size_in_bytes);
 218 
 219   bind(done);
 220 }
 221 
 222 
 223 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
 224   assert_different_registers(obj, t1, t2); // XXX really?
 225   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
 226 
 227   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
 228 
 229   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
 230 }
 231 
 232 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, bool is_tlab_allocated) {
 233   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
 234          "con_size_in_bytes is not multiple of alignment");
 235   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
 236 
 237   initialize_header(obj, klass, noreg, t1, t2);
 238 
 239   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
 240      // clear rest of allocated space
 241      const Register index = t2;
 242      const int threshold = 16 * BytesPerWord;   // approximate break even point for code size (see comments below)
 243      if (var_size_in_bytes != noreg) {
 244        mov(index, var_size_in_bytes);
 245        initialize_body(obj, index, hdr_size_in_bytes, t1);
 246      } else if (con_size_in_bytes <= threshold) {
 247        // use explicit null stores
 248        int i = hdr_size_in_bytes;
 249        if (i < con_size_in_bytes && (con_size_in_bytes % (2 * BytesPerWord))) {
 250          str(zr, Address(obj, i));
 251          i += BytesPerWord;
 252        }
 253        for (; i < con_size_in_bytes; i += 2 * BytesPerWord)
 254          stp(zr, zr, Address(obj, i));
 255      } else if (con_size_in_bytes > hdr_size_in_bytes) {
 256        block_comment("zero memory");
 257       // use loop to null out the fields
 258 
 259        int words = (con_size_in_bytes - hdr_size_in_bytes) / BytesPerWord;
 260        mov(index,  words / 8);
 261 
 262        const int unroll = 8; // Number of str(zr) instructions we'll unroll
 263        int remainder = words % unroll;
 264        lea(rscratch1, Address(obj, hdr_size_in_bytes + remainder * BytesPerWord));
 265 
 266        Label entry_point, loop;
 267        b(entry_point);
 268 
 269        bind(loop);
 270        sub(index, index, 1);
 271        for (int i = -unroll; i < 0; i++) {
 272          if (-i == remainder)
 273            bind(entry_point);
 274          str(zr, Address(rscratch1, i * wordSize));
 275        }
 276        if (remainder == 0)
 277          bind(entry_point);
 278        add(rscratch1, rscratch1, unroll * wordSize);
 279        cbnz(index, loop);
 280 
 281      }
 282   }
 283 
 284   membar(StoreStore);
 285 
 286   if (CURRENT_ENV->dtrace_alloc_probes()) {
 287     assert(obj == r0, "must be");
 288     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
 289   }
 290 
 291   verify_oop(obj);
 292 }
 293 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, int f, Register klass, Label& slow_case) {
 294   assert_different_registers(obj, len, t1, t2, klass);
 295 
 296   // determine alignment mask
 297   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
 298 
 299   // check for negative or excessive length
 300   mov(rscratch1, (int32_t)max_array_allocation_length);
 301   cmp(len, rscratch1);
 302   br(Assembler::HS, slow_case);
 303 
 304   const Register arr_size = t2; // okay to be the same
 305   // align object end
 306   mov(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
 307   add(arr_size, arr_size, len, ext::uxtw, f);
 308   andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
 309 
 310   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
 311 
 312   initialize_header(obj, klass, len, t1, t2);
 313 
 314   // clear rest of allocated space
 315   const Register len_zero = len;
 316   initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
 317 
 318   membar(StoreStore);
 319 
 320   if (CURRENT_ENV->dtrace_alloc_probes()) {
 321     assert(obj == r0, "must be");
 322     far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
 323   }
 324 
 325   verify_oop(obj);
 326 }
 327 
 328 
 329 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
 330   verify_oop(receiver);
 331   // explicit NULL check not needed since load from [klass_offset] causes a trap
 332   // check against inline cache
 333   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
 334 
 335   cmp_klass(receiver, iCache, rscratch1);
 336 }
 337 
 338 
 339 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes) {
 340   // If we have to make this method not-entrant we'll overwrite its
 341   // first instruction with a jump.  For this action to be legal we
 342   // must ensure that this first instruction is a B, BL, NOP, BKPT,
 343   // SVC, HVC, or SMC.  Make it a NOP.
 344   nop();
 345   assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
 346   // Make sure there is enough stack space for this method's activation.
 347   // Note that we do this before doing an enter().
 348   generate_stack_overflow_check(bang_size_in_bytes);
 349   MacroAssembler::build_frame(framesize + 2 * wordSize);
 350   if (NotifySimulator) {
 351     notify(Assembler::method_entry);
 352   }
 353 }
 354 
 355 void C1_MacroAssembler::remove_frame(int framesize) {
 356   MacroAssembler::remove_frame(framesize + 2 * wordSize);
 357   if (NotifySimulator) {
 358     notify(Assembler::method_reentry);
 359   }
 360 }
 361 
 362 
 363 void C1_MacroAssembler::verified_entry() {
 364 }
 365 
 366 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
 367   // rbp, + 0: link
 368   //     + 1: return address
 369   //     + 2: argument with offset 0
 370   //     + 3: argument with offset 1
 371   //     + 4: ...
 372 
 373   ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
 374 }
 375 
 376 #ifndef PRODUCT
 377 
 378 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
 379   if (!VerifyOops) return;
 380   verify_oop_addr(Address(sp, stack_offset), "oop");
 381 }
 382 
 383 void C1_MacroAssembler::verify_not_null_oop(Register r) {
 384   if (!VerifyOops) return;
 385   Label not_null;
 386   cbnz(r, not_null);
 387   stop("non-null oop required");
 388   bind(not_null);
 389   verify_oop(r);
 390 }
 391 
 392 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
 393 #ifdef ASSERT
 394   static int nn;
 395   if (inv_r0) mov(r0, 0xDEAD);
 396   if (inv_r19) mov(r19, 0xDEAD);
 397   if (inv_r2) mov(r2, nn++);
 398   if (inv_r3) mov(r3, 0xDEAD);
 399   if (inv_r4) mov(r4, 0xDEAD);
 400   if (inv_r5) mov(r5, 0xDEAD);
 401 #endif
 402 }
 403 #endif // ifndef PRODUCT
--- EOF ---