Mercurial > hg > truffle
annotate src/cpu/x86/vm/stubGenerator_x86_64.cpp @ 356:1ee8caae33af
Merge
| author | tonyp |
|---|---|
| date | Thu, 21 Aug 2008 23:36:31 -0400 |
| parents | 37f87013dfd8 910a4cb98e9e |
| children | f8199438385b |
| rev | line source |
|---|---|
| 0 | 1 /* |
| 196 | 2 * Copyright 2003-2008 Sun Microsystems, Inc. All Rights Reserved. |
| 0 | 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
| 4 * | |
| 5 * This code is free software; you can redistribute it and/or modify it | |
| 6 * under the terms of the GNU General Public License version 2 only, as | |
| 7 * published by the Free Software Foundation. | |
| 8 * | |
| 9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
| 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
| 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
| 12 * version 2 for more details (a copy is included in the LICENSE file that | |
| 13 * accompanied this code). | |
| 14 * | |
| 15 * You should have received a copy of the GNU General Public License version | |
| 16 * 2 along with this work; if not, write to the Free Software Foundation, | |
| 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
| 18 * | |
| 19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
| 20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
| 21 * have any questions. | |
| 22 * | |
| 23 */ | |
| 24 | |
| 25 #include "incls/_precompiled.incl" | |
| 26 #include "incls/_stubGenerator_x86_64.cpp.incl" | |
| 27 | |
| 28 // Declaration and definition of StubGenerator (no .hpp file). | |
| 29 // For a more detailed description of the stub routine structure | |
| 30 // see the comment in stubRoutines.hpp | |
| 31 | |
| 32 #define __ _masm-> | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
33 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8) |
| 0 | 34 |
| 35 #ifdef PRODUCT | |
| 36 #define BLOCK_COMMENT(str) /* nothing */ | |
| 37 #else | |
| 38 #define BLOCK_COMMENT(str) __ block_comment(str) | |
| 39 #endif | |
| 40 | |
| 41 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") | |
| 42 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions | |
| 43 | |
| 44 // Stub Code definitions | |
| 45 | |
| 46 static address handle_unsafe_access() { | |
| 47 JavaThread* thread = JavaThread::current(); | |
| 48 address pc = thread->saved_exception_pc(); | |
| 49 // pc is the instruction which we must emulate | |
| 50 // doing a no-op is fine: return garbage from the load | |
| 51 // therefore, compute npc | |
| 52 address npc = Assembler::locate_next_instruction(pc); | |
| 53 | |
| 54 // request an async exception | |
| 55 thread->set_pending_unsafe_access_error(); | |
| 56 | |
| 57 // return address of next instruction to execute | |
| 58 return npc; | |
| 59 } | |
| 60 | |
| 61 class StubGenerator: public StubCodeGenerator { | |
| 62 private: | |
| 63 | |
| 64 #ifdef PRODUCT | |
| 65 #define inc_counter_np(counter) (0) | |
| 66 #else | |
| 67 void inc_counter_np_(int& counter) { | |
| 68 __ incrementl(ExternalAddress((address)&counter)); | |
| 69 } | |
| 70 #define inc_counter_np(counter) \ | |
| 71 BLOCK_COMMENT("inc_counter " #counter); \ | |
| 72 inc_counter_np_(counter); | |
| 73 #endif | |
| 74 | |
| 75 // Call stubs are used to call Java from C | |
| 76 // | |
| 77 // Linux Arguments: | |
| 78 // c_rarg0: call wrapper address address | |
| 79 // c_rarg1: result address | |
| 80 // c_rarg2: result type BasicType | |
| 81 // c_rarg3: method methodOop | |
| 82 // c_rarg4: (interpreter) entry point address | |
| 83 // c_rarg5: parameters intptr_t* | |
| 84 // 16(rbp): parameter size (in words) int | |
| 85 // 24(rbp): thread Thread* | |
| 86 // | |
| 87 // [ return_from_Java ] <--- rsp | |
| 88 // [ argument word n ] | |
| 89 // ... | |
| 90 // -12 [ argument word 1 ] | |
| 91 // -11 [ saved r15 ] <--- rsp_after_call | |
| 92 // -10 [ saved r14 ] | |
| 93 // -9 [ saved r13 ] | |
| 94 // -8 [ saved r12 ] | |
| 95 // -7 [ saved rbx ] | |
| 96 // -6 [ call wrapper ] | |
| 97 // -5 [ result ] | |
| 98 // -4 [ result type ] | |
| 99 // -3 [ method ] | |
| 100 // -2 [ entry point ] | |
| 101 // -1 [ parameters ] | |
| 102 // 0 [ saved rbp ] <--- rbp | |
| 103 // 1 [ return address ] | |
| 104 // 2 [ parameter size ] | |
| 105 // 3 [ thread ] | |
| 106 // | |
| 107 // Windows Arguments: | |
| 108 // c_rarg0: call wrapper address address | |
| 109 // c_rarg1: result address | |
| 110 // c_rarg2: result type BasicType | |
| 111 // c_rarg3: method methodOop | |
| 112 // 48(rbp): (interpreter) entry point address | |
| 113 // 56(rbp): parameters intptr_t* | |
| 114 // 64(rbp): parameter size (in words) int | |
| 115 // 72(rbp): thread Thread* | |
| 116 // | |
| 117 // [ return_from_Java ] <--- rsp | |
| 118 // [ argument word n ] | |
| 119 // ... | |
| 120 // -8 [ argument word 1 ] | |
| 121 // -7 [ saved r15 ] <--- rsp_after_call | |
| 122 // -6 [ saved r14 ] | |
| 123 // -5 [ saved r13 ] | |
| 124 // -4 [ saved r12 ] | |
| 125 // -3 [ saved rdi ] | |
| 126 // -2 [ saved rsi ] | |
| 127 // -1 [ saved rbx ] | |
| 128 // 0 [ saved rbp ] <--- rbp | |
| 129 // 1 [ return address ] | |
| 130 // 2 [ call wrapper ] | |
| 131 // 3 [ result ] | |
| 132 // 4 [ result type ] | |
| 133 // 5 [ method ] | |
| 134 // 6 [ entry point ] | |
| 135 // 7 [ parameters ] | |
| 136 // 8 [ parameter size ] | |
| 137 // 9 [ thread ] | |
| 138 // | |
| 139 // Windows reserves the callers stack space for arguments 1-4. | |
| 140 // We spill c_rarg0-c_rarg3 to this space. | |
| 141 | |
| 142 // Call stub stack layout word offsets from rbp | |
| 143 enum call_stub_layout { | |
| 144 #ifdef _WIN64 | |
| 145 rsp_after_call_off = -7, | |
| 146 r15_off = rsp_after_call_off, | |
| 147 r14_off = -6, | |
| 148 r13_off = -5, | |
| 149 r12_off = -4, | |
| 150 rdi_off = -3, | |
| 151 rsi_off = -2, | |
| 152 rbx_off = -1, | |
| 153 rbp_off = 0, | |
| 154 retaddr_off = 1, | |
| 155 call_wrapper_off = 2, | |
| 156 result_off = 3, | |
| 157 result_type_off = 4, | |
| 158 method_off = 5, | |
| 159 entry_point_off = 6, | |
| 160 parameters_off = 7, | |
| 161 parameter_size_off = 8, | |
| 162 thread_off = 9 | |
| 163 #else | |
| 164 rsp_after_call_off = -12, | |
| 165 mxcsr_off = rsp_after_call_off, | |
| 166 r15_off = -11, | |
| 167 r14_off = -10, | |
| 168 r13_off = -9, | |
| 169 r12_off = -8, | |
| 170 rbx_off = -7, | |
| 171 call_wrapper_off = -6, | |
| 172 result_off = -5, | |
| 173 result_type_off = -4, | |
| 174 method_off = -3, | |
| 175 entry_point_off = -2, | |
| 176 parameters_off = -1, | |
| 177 rbp_off = 0, | |
| 178 retaddr_off = 1, | |
| 179 parameter_size_off = 2, | |
| 180 thread_off = 3 | |
| 181 #endif | |
| 182 }; | |
| 183 | |
| 184 address generate_call_stub(address& return_address) { | |
| 185 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 && | |
| 186 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off, | |
| 187 "adjust this code"); | |
| 188 StubCodeMark mark(this, "StubRoutines", "call_stub"); | |
| 189 address start = __ pc(); | |
| 190 | |
| 191 // same as in generate_catch_exception()! | |
| 192 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize); | |
| 193 | |
| 194 const Address call_wrapper (rbp, call_wrapper_off * wordSize); | |
| 195 const Address result (rbp, result_off * wordSize); | |
| 196 const Address result_type (rbp, result_type_off * wordSize); | |
| 197 const Address method (rbp, method_off * wordSize); | |
| 198 const Address entry_point (rbp, entry_point_off * wordSize); | |
| 199 const Address parameters (rbp, parameters_off * wordSize); | |
| 200 const Address parameter_size(rbp, parameter_size_off * wordSize); | |
| 201 | |
| 202 // same as in generate_catch_exception()! | |
| 203 const Address thread (rbp, thread_off * wordSize); | |
| 204 | |
| 205 const Address r15_save(rbp, r15_off * wordSize); | |
| 206 const Address r14_save(rbp, r14_off * wordSize); | |
| 207 const Address r13_save(rbp, r13_off * wordSize); | |
| 208 const Address r12_save(rbp, r12_off * wordSize); | |
| 209 const Address rbx_save(rbp, rbx_off * wordSize); | |
| 210 | |
| 211 // stub code | |
| 212 __ enter(); | |
| 213 __ subq(rsp, -rsp_after_call_off * wordSize); | |
| 214 | |
| 215 // save register parameters | |
| 216 #ifndef _WIN64 | |
| 217 __ movq(parameters, c_rarg5); // parameters | |
| 218 __ movq(entry_point, c_rarg4); // entry_point | |
| 219 #endif | |
| 220 | |
| 221 __ movq(method, c_rarg3); // method | |
| 222 __ movl(result_type, c_rarg2); // result type | |
| 223 __ movq(result, c_rarg1); // result | |
| 224 __ movq(call_wrapper, c_rarg0); // call wrapper | |
| 225 | |
| 226 // save regs belonging to calling function | |
| 227 __ movq(rbx_save, rbx); | |
| 228 __ movq(r12_save, r12); | |
| 229 __ movq(r13_save, r13); | |
| 230 __ movq(r14_save, r14); | |
| 231 __ movq(r15_save, r15); | |
| 232 | |
| 233 #ifdef _WIN64 | |
| 234 const Address rdi_save(rbp, rdi_off * wordSize); | |
| 235 const Address rsi_save(rbp, rsi_off * wordSize); | |
| 236 | |
| 237 __ movq(rsi_save, rsi); | |
| 238 __ movq(rdi_save, rdi); | |
| 239 #else | |
| 240 const Address mxcsr_save(rbp, mxcsr_off * wordSize); | |
| 241 { | |
| 242 Label skip_ldmx; | |
| 243 __ stmxcsr(mxcsr_save); | |
| 244 __ movl(rax, mxcsr_save); | |
| 245 __ andl(rax, MXCSR_MASK); // Only check control and mask bits | |
| 246 ExternalAddress mxcsr_std(StubRoutines::amd64::mxcsr_std()); | |
| 247 __ cmp32(rax, mxcsr_std); | |
| 248 __ jcc(Assembler::equal, skip_ldmx); | |
| 249 __ ldmxcsr(mxcsr_std); | |
| 250 __ bind(skip_ldmx); | |
| 251 } | |
| 252 #endif | |
| 253 | |
| 254 // Load up thread register | |
| 255 __ movq(r15_thread, thread); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
256 __ reinit_heapbase(); |
| 0 | 257 |
| 258 #ifdef ASSERT | |
| 259 // make sure we have no pending exceptions | |
| 260 { | |
| 261 Label L; | |
| 262 __ cmpq(Address(r15_thread, Thread::pending_exception_offset()), (int)NULL_WORD); | |
| 263 __ jcc(Assembler::equal, L); | |
| 264 __ stop("StubRoutines::call_stub: entered with pending exception"); | |
| 265 __ bind(L); | |
| 266 } | |
| 267 #endif | |
| 268 | |
| 269 // pass parameters if any | |
| 270 BLOCK_COMMENT("pass parameters if any"); | |
| 271 Label parameters_done; | |
| 272 __ movl(c_rarg3, parameter_size); | |
| 273 __ testl(c_rarg3, c_rarg3); | |
| 274 __ jcc(Assembler::zero, parameters_done); | |
| 275 | |
| 276 Label loop; | |
| 277 __ movq(c_rarg2, parameters); // parameter pointer | |
| 278 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1 | |
| 279 __ BIND(loop); | |
| 280 if (TaggedStackInterpreter) { | |
| 281 __ movq(rax, Address(c_rarg2, 0)); // get tag | |
| 282 __ addq(c_rarg2, wordSize); // advance to next tag | |
| 283 __ pushq(rax); // pass tag | |
| 284 } | |
| 285 __ movq(rax, Address(c_rarg2, 0)); // get parameter | |
| 286 __ addq(c_rarg2, wordSize); // advance to next parameter | |
| 287 __ decrementl(c_rarg1); // decrement counter | |
| 288 __ pushq(rax); // pass parameter | |
| 289 __ jcc(Assembler::notZero, loop); | |
| 290 | |
| 291 // call Java function | |
| 292 __ BIND(parameters_done); | |
| 293 __ movq(rbx, method); // get methodOop | |
| 294 __ movq(c_rarg1, entry_point); // get entry_point | |
| 295 __ movq(r13, rsp); // set sender sp | |
| 296 BLOCK_COMMENT("call Java function"); | |
| 297 __ call(c_rarg1); | |
| 298 | |
| 299 BLOCK_COMMENT("call_stub_return_address:"); | |
| 300 return_address = __ pc(); | |
| 301 | |
| 302 // store result depending on type (everything that is not | |
| 303 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT) | |
| 304 __ movq(c_rarg0, result); | |
| 305 Label is_long, is_float, is_double, exit; | |
| 306 __ movl(c_rarg1, result_type); | |
| 307 __ cmpl(c_rarg1, T_OBJECT); | |
| 308 __ jcc(Assembler::equal, is_long); | |
| 309 __ cmpl(c_rarg1, T_LONG); | |
| 310 __ jcc(Assembler::equal, is_long); | |
| 311 __ cmpl(c_rarg1, T_FLOAT); | |
| 312 __ jcc(Assembler::equal, is_float); | |
| 313 __ cmpl(c_rarg1, T_DOUBLE); | |
| 314 __ jcc(Assembler::equal, is_double); | |
| 315 | |
| 316 // handle T_INT case | |
| 317 __ movl(Address(c_rarg0, 0), rax); | |
| 318 | |
| 319 __ BIND(exit); | |
| 320 | |
| 321 // pop parameters | |
| 322 __ leaq(rsp, rsp_after_call); | |
| 323 | |
| 324 #ifdef ASSERT | |
| 325 // verify that threads correspond | |
| 326 { | |
| 327 Label L, S; | |
| 328 __ cmpq(r15_thread, thread); | |
| 329 __ jcc(Assembler::notEqual, S); | |
| 330 __ get_thread(rbx); | |
| 331 __ cmpq(r15_thread, rbx); | |
| 332 __ jcc(Assembler::equal, L); | |
| 333 __ bind(S); | |
| 334 __ jcc(Assembler::equal, L); | |
| 335 __ stop("StubRoutines::call_stub: threads must correspond"); | |
| 336 __ bind(L); | |
| 337 } | |
| 338 #endif | |
| 339 | |
| 340 // restore regs belonging to calling function | |
| 341 __ movq(r15, r15_save); | |
| 342 __ movq(r14, r14_save); | |
| 343 __ movq(r13, r13_save); | |
| 344 __ movq(r12, r12_save); | |
| 345 __ movq(rbx, rbx_save); | |
| 346 | |
| 347 #ifdef _WIN64 | |
| 348 __ movq(rdi, rdi_save); | |
| 349 __ movq(rsi, rsi_save); | |
| 350 #else | |
| 351 __ ldmxcsr(mxcsr_save); | |
| 352 #endif | |
| 353 | |
| 354 // restore rsp | |
| 355 __ addq(rsp, -rsp_after_call_off * wordSize); | |
| 356 | |
| 357 // return | |
| 358 __ popq(rbp); | |
| 359 __ ret(0); | |
| 360 | |
| 361 // handle return types different from T_INT | |
| 362 __ BIND(is_long); | |
| 363 __ movq(Address(c_rarg0, 0), rax); | |
| 364 __ jmp(exit); | |
| 365 | |
| 366 __ BIND(is_float); | |
| 367 __ movflt(Address(c_rarg0, 0), xmm0); | |
| 368 __ jmp(exit); | |
| 369 | |
| 370 __ BIND(is_double); | |
| 371 __ movdbl(Address(c_rarg0, 0), xmm0); | |
| 372 __ jmp(exit); | |
| 373 | |
| 374 return start; | |
| 375 } | |
| 376 | |
| 377 // Return point for a Java call if there's an exception thrown in | |
| 378 // Java code. The exception is caught and transformed into a | |
| 379 // pending exception stored in JavaThread that can be tested from | |
| 380 // within the VM. | |
| 381 // | |
| 382 // Note: Usually the parameters are removed by the callee. In case | |
| 383 // of an exception crossing an activation frame boundary, that is | |
| 384 // not the case if the callee is compiled code => need to setup the | |
| 385 // rsp. | |
| 386 // | |
| 387 // rax: exception oop | |
| 388 | |
| 389 address generate_catch_exception() { | |
| 390 StubCodeMark mark(this, "StubRoutines", "catch_exception"); | |
| 391 address start = __ pc(); | |
| 392 | |
| 393 // same as in generate_call_stub(): | |
| 394 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize); | |
| 395 const Address thread (rbp, thread_off * wordSize); | |
| 396 | |
| 397 #ifdef ASSERT | |
| 398 // verify that threads correspond | |
| 399 { | |
| 400 Label L, S; | |
| 401 __ cmpq(r15_thread, thread); | |
| 402 __ jcc(Assembler::notEqual, S); | |
| 403 __ get_thread(rbx); | |
| 404 __ cmpq(r15_thread, rbx); | |
| 405 __ jcc(Assembler::equal, L); | |
| 406 __ bind(S); | |
| 407 __ stop("StubRoutines::catch_exception: threads must correspond"); | |
| 408 __ bind(L); | |
| 409 } | |
| 410 #endif | |
| 411 | |
| 412 // set pending exception | |
| 413 __ verify_oop(rax); | |
| 414 | |
| 415 __ movq(Address(r15_thread, Thread::pending_exception_offset()), rax); | |
| 416 __ lea(rscratch1, ExternalAddress((address)__FILE__)); | |
| 417 __ movq(Address(r15_thread, Thread::exception_file_offset()), rscratch1); | |
| 418 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__); | |
| 419 | |
| 420 // complete return to VM | |
| 421 assert(StubRoutines::_call_stub_return_address != NULL, | |
| 422 "_call_stub_return_address must have been generated before"); | |
| 423 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address)); | |
| 424 | |
| 425 return start; | |
| 426 } | |
| 427 | |
| 428 // Continuation point for runtime calls returning with a pending | |
| 429 // exception. The pending exception check happened in the runtime | |
| 430 // or native call stub. The pending exception in Thread is | |
| 431 // converted into a Java-level exception. | |
| 432 // | |
| 433 // Contract with Java-level exception handlers: | |
| 434 // rax: exception | |
| 435 // rdx: throwing pc | |
| 436 // | |
| 437 // NOTE: At entry of this stub, exception-pc must be on stack !! | |
| 438 | |
| 439 address generate_forward_exception() { | |
| 440 StubCodeMark mark(this, "StubRoutines", "forward exception"); | |
| 441 address start = __ pc(); | |
| 442 | |
| 443 // Upon entry, the sp points to the return address returning into | |
| 444 // Java (interpreted or compiled) code; i.e., the return address | |
| 445 // becomes the throwing pc. | |
| 446 // | |
| 447 // Arguments pushed before the runtime call are still on the stack | |
| 448 // but the exception handler will reset the stack pointer -> | |
| 449 // ignore them. A potential result in registers can be ignored as | |
| 450 // well. | |
| 451 | |
| 452 #ifdef ASSERT | |
| 453 // make sure this code is only executed if there is a pending exception | |
| 454 { | |
| 455 Label L; | |
| 456 __ cmpq(Address(r15_thread, Thread::pending_exception_offset()), (int) NULL); | |
| 457 __ jcc(Assembler::notEqual, L); | |
| 458 __ stop("StubRoutines::forward exception: no pending exception (1)"); | |
| 459 __ bind(L); | |
| 460 } | |
| 461 #endif | |
| 462 | |
| 463 // compute exception handler into rbx | |
| 464 __ movq(c_rarg0, Address(rsp, 0)); | |
| 465 BLOCK_COMMENT("call exception_handler_for_return_address"); | |
| 466 __ call_VM_leaf(CAST_FROM_FN_PTR(address, | |
| 467 SharedRuntime::exception_handler_for_return_address), | |
| 468 c_rarg0); | |
| 469 __ movq(rbx, rax); | |
| 470 | |
| 471 // setup rax & rdx, remove return address & clear pending exception | |
| 472 __ popq(rdx); | |
| 473 __ movq(rax, Address(r15_thread, Thread::pending_exception_offset())); | |
| 474 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int)NULL_WORD); | |
| 475 | |
| 476 #ifdef ASSERT | |
| 477 // make sure exception is set | |
| 478 { | |
| 479 Label L; | |
| 480 __ testq(rax, rax); | |
| 481 __ jcc(Assembler::notEqual, L); | |
| 482 __ stop("StubRoutines::forward exception: no pending exception (2)"); | |
| 483 __ bind(L); | |
| 484 } | |
| 485 #endif | |
| 486 | |
| 487 // continue at exception handler (return address removed) | |
| 488 // rax: exception | |
| 489 // rbx: exception handler | |
| 490 // rdx: throwing pc | |
| 491 __ verify_oop(rax); | |
| 492 __ jmp(rbx); | |
| 493 | |
| 494 return start; | |
| 495 } | |
| 496 | |
| 497 // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest) | |
| 498 // | |
| 499 // Arguments : | |
| 500 // c_rarg0: exchange_value | |
| 501 // c_rarg0: dest | |
| 502 // | |
| 503 // Result: | |
| 504 // *dest <- ex, return (orig *dest) | |
| 505 address generate_atomic_xchg() { | |
| 506 StubCodeMark mark(this, "StubRoutines", "atomic_xchg"); | |
| 507 address start = __ pc(); | |
| 508 | |
| 509 __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow | |
| 510 __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK | |
| 511 __ ret(0); | |
| 512 | |
| 513 return start; | |
| 514 } | |
| 515 | |
| 516 // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest) | |
| 517 // | |
| 518 // Arguments : | |
| 519 // c_rarg0: exchange_value | |
| 520 // c_rarg1: dest | |
| 521 // | |
| 522 // Result: | |
| 523 // *dest <- ex, return (orig *dest) | |
| 524 address generate_atomic_xchg_ptr() { | |
| 525 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr"); | |
| 526 address start = __ pc(); | |
| 527 | |
| 528 __ movq(rax, c_rarg0); // Copy to eax we need a return value anyhow | |
| 529 __ xchgq(rax, Address(c_rarg1, 0)); // automatic LOCK | |
| 530 __ ret(0); | |
| 531 | |
| 532 return start; | |
| 533 } | |
| 534 | |
| 535 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest, | |
| 536 // jint compare_value) | |
| 537 // | |
| 538 // Arguments : | |
| 539 // c_rarg0: exchange_value | |
| 540 // c_rarg1: dest | |
| 541 // c_rarg2: compare_value | |
| 542 // | |
| 543 // Result: | |
| 544 // if ( compare_value == *dest ) { | |
| 545 // *dest = exchange_value | |
| 546 // return compare_value; | |
| 547 // else | |
| 548 // return *dest; | |
| 549 address generate_atomic_cmpxchg() { | |
| 550 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg"); | |
| 551 address start = __ pc(); | |
| 552 | |
| 553 __ movl(rax, c_rarg2); | |
| 554 if ( os::is_MP() ) __ lock(); | |
| 555 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0)); | |
| 556 __ ret(0); | |
| 557 | |
| 558 return start; | |
| 559 } | |
| 560 | |
| 561 // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value, | |
| 562 // volatile jlong* dest, | |
| 563 // jlong compare_value) | |
| 564 // Arguments : | |
| 565 // c_rarg0: exchange_value | |
| 566 // c_rarg1: dest | |
| 567 // c_rarg2: compare_value | |
| 568 // | |
| 569 // Result: | |
| 570 // if ( compare_value == *dest ) { | |
| 571 // *dest = exchange_value | |
| 572 // return compare_value; | |
| 573 // else | |
| 574 // return *dest; | |
| 575 address generate_atomic_cmpxchg_long() { | |
| 576 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long"); | |
| 577 address start = __ pc(); | |
| 578 | |
| 579 __ movq(rax, c_rarg2); | |
| 580 if ( os::is_MP() ) __ lock(); | |
| 581 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0)); | |
| 582 __ ret(0); | |
| 583 | |
| 584 return start; | |
| 585 } | |
| 586 | |
| 587 // Support for jint atomic::add(jint add_value, volatile jint* dest) | |
| 588 // | |
| 589 // Arguments : | |
| 590 // c_rarg0: add_value | |
| 591 // c_rarg1: dest | |
| 592 // | |
| 593 // Result: | |
| 594 // *dest += add_value | |
| 595 // return *dest; | |
| 596 address generate_atomic_add() { | |
| 597 StubCodeMark mark(this, "StubRoutines", "atomic_add"); | |
| 598 address start = __ pc(); | |
| 599 | |
| 600 __ movl(rax, c_rarg0); | |
| 601 if ( os::is_MP() ) __ lock(); | |
| 602 __ xaddl(Address(c_rarg1, 0), c_rarg0); | |
| 603 __ addl(rax, c_rarg0); | |
| 604 __ ret(0); | |
| 605 | |
| 606 return start; | |
| 607 } | |
| 608 | |
| 609 // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest) | |
| 610 // | |
| 611 // Arguments : | |
| 612 // c_rarg0: add_value | |
| 613 // c_rarg1: dest | |
| 614 // | |
| 615 // Result: | |
| 616 // *dest += add_value | |
| 617 // return *dest; | |
| 618 address generate_atomic_add_ptr() { | |
| 619 StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr"); | |
| 620 address start = __ pc(); | |
| 621 | |
| 622 __ movq(rax, c_rarg0); // Copy to eax we need a return value anyhow | |
| 623 if ( os::is_MP() ) __ lock(); | |
| 624 __ xaddl(Address(c_rarg1, 0), c_rarg0); | |
| 625 __ addl(rax, c_rarg0); | |
| 626 __ ret(0); | |
| 627 | |
| 628 return start; | |
| 629 } | |
| 630 | |
| 631 // Support for intptr_t OrderAccess::fence() | |
| 632 // | |
| 633 // Arguments : | |
| 634 // | |
| 635 // Result: | |
| 636 address generate_orderaccess_fence() { | |
| 637 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence"); | |
| 638 address start = __ pc(); | |
| 639 __ mfence(); | |
| 640 __ ret(0); | |
| 641 | |
| 642 return start; | |
| 643 } | |
| 644 | |
| 645 // Support for intptr_t get_previous_fp() | |
| 646 // | |
| 647 // This routine is used to find the previous frame pointer for the | |
| 648 // caller (current_frame_guess). This is used as part of debugging | |
| 649 // ps() is seemingly lost trying to find frames. | |
| 650 // This code assumes that caller current_frame_guess) has a frame. | |
| 651 address generate_get_previous_fp() { | |
| 652 StubCodeMark mark(this, "StubRoutines", "get_previous_fp"); | |
| 653 const Address old_fp(rbp, 0); | |
| 654 const Address older_fp(rax, 0); | |
| 655 address start = __ pc(); | |
| 656 | |
| 657 __ enter(); | |
| 658 __ movq(rax, old_fp); // callers fp | |
| 659 __ movq(rax, older_fp); // the frame for ps() | |
| 660 __ popq(rbp); | |
| 661 __ ret(0); | |
| 662 | |
| 663 return start; | |
| 664 } | |
| 665 | |
| 666 //---------------------------------------------------------------------------------------------------- | |
| 667 // Support for void verify_mxcsr() | |
| 668 // | |
| 669 // This routine is used with -Xcheck:jni to verify that native | |
| 670 // JNI code does not return to Java code without restoring the | |
| 671 // MXCSR register to our expected state. | |
| 672 | |
| 673 address generate_verify_mxcsr() { | |
| 674 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr"); | |
| 675 address start = __ pc(); | |
| 676 | |
| 677 const Address mxcsr_save(rsp, 0); | |
| 678 | |
| 679 if (CheckJNICalls) { | |
| 680 Label ok_ret; | |
| 681 __ pushq(rax); | |
| 682 __ subq(rsp, wordSize); // allocate a temp location | |
| 683 __ stmxcsr(mxcsr_save); | |
| 684 __ movl(rax, mxcsr_save); | |
| 685 __ andl(rax, MXCSR_MASK); // Only check control and mask bits | |
| 686 __ cmpl(rax, *(int *)(StubRoutines::amd64::mxcsr_std())); | |
| 687 __ jcc(Assembler::equal, ok_ret); | |
| 688 | |
| 689 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall"); | |
| 690 | |
| 691 __ ldmxcsr(ExternalAddress(StubRoutines::amd64::mxcsr_std())); | |
| 692 | |
| 693 __ bind(ok_ret); | |
| 694 __ addq(rsp, wordSize); | |
| 695 __ popq(rax); | |
| 696 } | |
| 697 | |
| 698 __ ret(0); | |
| 699 | |
| 700 return start; | |
| 701 } | |
| 702 | |
| 703 address generate_f2i_fixup() { | |
| 704 StubCodeMark mark(this, "StubRoutines", "f2i_fixup"); | |
| 705 Address inout(rsp, 5 * wordSize); // return address + 4 saves | |
| 706 | |
| 707 address start = __ pc(); | |
| 708 | |
| 709 Label L; | |
| 710 | |
| 711 __ pushq(rax); | |
| 712 __ pushq(c_rarg3); | |
| 713 __ pushq(c_rarg2); | |
| 714 __ pushq(c_rarg1); | |
| 715 | |
| 716 __ movl(rax, 0x7f800000); | |
| 717 __ xorl(c_rarg3, c_rarg3); | |
| 718 __ movl(c_rarg2, inout); | |
| 719 __ movl(c_rarg1, c_rarg2); | |
| 720 __ andl(c_rarg1, 0x7fffffff); | |
| 721 __ cmpl(rax, c_rarg1); // NaN? -> 0 | |
| 722 __ jcc(Assembler::negative, L); | |
| 723 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint | |
| 724 __ movl(c_rarg3, 0x80000000); | |
| 725 __ movl(rax, 0x7fffffff); | |
| 726 __ cmovl(Assembler::positive, c_rarg3, rax); | |
| 727 | |
| 728 __ bind(L); | |
| 729 __ movq(inout, c_rarg3); | |
| 730 | |
| 731 __ popq(c_rarg1); | |
| 732 __ popq(c_rarg2); | |
| 733 __ popq(c_rarg3); | |
| 734 __ popq(rax); | |
| 735 | |
| 736 __ ret(0); | |
| 737 | |
| 738 return start; | |
| 739 } | |
| 740 | |
| 741 address generate_f2l_fixup() { | |
| 742 StubCodeMark mark(this, "StubRoutines", "f2l_fixup"); | |
| 743 Address inout(rsp, 5 * wordSize); // return address + 4 saves | |
| 744 address start = __ pc(); | |
| 745 | |
| 746 Label L; | |
| 747 | |
| 748 __ pushq(rax); | |
| 749 __ pushq(c_rarg3); | |
| 750 __ pushq(c_rarg2); | |
| 751 __ pushq(c_rarg1); | |
| 752 | |
| 753 __ movl(rax, 0x7f800000); | |
| 754 __ xorl(c_rarg3, c_rarg3); | |
| 755 __ movl(c_rarg2, inout); | |
| 756 __ movl(c_rarg1, c_rarg2); | |
| 757 __ andl(c_rarg1, 0x7fffffff); | |
| 758 __ cmpl(rax, c_rarg1); // NaN? -> 0 | |
| 759 __ jcc(Assembler::negative, L); | |
| 760 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong | |
| 761 __ mov64(c_rarg3, 0x8000000000000000); | |
| 762 __ mov64(rax, 0x7fffffffffffffff); | |
| 763 __ cmovq(Assembler::positive, c_rarg3, rax); | |
| 764 | |
| 765 __ bind(L); | |
| 766 __ movq(inout, c_rarg3); | |
| 767 | |
| 768 __ popq(c_rarg1); | |
| 769 __ popq(c_rarg2); | |
| 770 __ popq(c_rarg3); | |
| 771 __ popq(rax); | |
| 772 | |
| 773 __ ret(0); | |
| 774 | |
| 775 return start; | |
| 776 } | |
| 777 | |
| 778 address generate_d2i_fixup() { | |
| 779 StubCodeMark mark(this, "StubRoutines", "d2i_fixup"); | |
| 780 Address inout(rsp, 6 * wordSize); // return address + 5 saves | |
| 781 | |
| 782 address start = __ pc(); | |
| 783 | |
| 784 Label L; | |
| 785 | |
| 786 __ pushq(rax); | |
| 787 __ pushq(c_rarg3); | |
| 788 __ pushq(c_rarg2); | |
| 789 __ pushq(c_rarg1); | |
| 790 __ pushq(c_rarg0); | |
| 791 | |
| 792 __ movl(rax, 0x7ff00000); | |
| 793 __ movq(c_rarg2, inout); | |
| 794 __ movl(c_rarg3, c_rarg2); | |
| 795 __ movq(c_rarg1, c_rarg2); | |
| 796 __ movq(c_rarg0, c_rarg2); | |
| 797 __ negl(c_rarg3); | |
| 798 __ shrq(c_rarg1, 0x20); | |
| 799 __ orl(c_rarg3, c_rarg2); | |
| 800 __ andl(c_rarg1, 0x7fffffff); | |
| 801 __ xorl(c_rarg2, c_rarg2); | |
| 802 __ shrl(c_rarg3, 0x1f); | |
| 803 __ orl(c_rarg1, c_rarg3); | |
| 804 __ cmpl(rax, c_rarg1); | |
| 805 __ jcc(Assembler::negative, L); // NaN -> 0 | |
| 806 __ testq(c_rarg0, c_rarg0); // signed ? min_jint : max_jint | |
| 807 __ movl(c_rarg2, 0x80000000); | |
| 808 __ movl(rax, 0x7fffffff); | |
| 809 __ cmovl(Assembler::positive, c_rarg2, rax); | |
| 810 | |
| 811 __ bind(L); | |
| 812 __ movq(inout, c_rarg2); | |
| 813 | |
| 814 __ popq(c_rarg0); | |
| 815 __ popq(c_rarg1); | |
| 816 __ popq(c_rarg2); | |
| 817 __ popq(c_rarg3); | |
| 818 __ popq(rax); | |
| 819 | |
| 820 __ ret(0); | |
| 821 | |
| 822 return start; | |
| 823 } | |
| 824 | |
| 825 address generate_d2l_fixup() { | |
| 826 StubCodeMark mark(this, "StubRoutines", "d2l_fixup"); | |
| 827 Address inout(rsp, 6 * wordSize); // return address + 5 saves | |
| 828 | |
| 829 address start = __ pc(); | |
| 830 | |
| 831 Label L; | |
| 832 | |
| 833 __ pushq(rax); | |
| 834 __ pushq(c_rarg3); | |
| 835 __ pushq(c_rarg2); | |
| 836 __ pushq(c_rarg1); | |
| 837 __ pushq(c_rarg0); | |
| 838 | |
| 839 __ movl(rax, 0x7ff00000); | |
| 840 __ movq(c_rarg2, inout); | |
| 841 __ movl(c_rarg3, c_rarg2); | |
| 842 __ movq(c_rarg1, c_rarg2); | |
| 843 __ movq(c_rarg0, c_rarg2); | |
| 844 __ negl(c_rarg3); | |
| 845 __ shrq(c_rarg1, 0x20); | |
| 846 __ orl(c_rarg3, c_rarg2); | |
| 847 __ andl(c_rarg1, 0x7fffffff); | |
| 848 __ xorl(c_rarg2, c_rarg2); | |
| 849 __ shrl(c_rarg3, 0x1f); | |
| 850 __ orl(c_rarg1, c_rarg3); | |
| 851 __ cmpl(rax, c_rarg1); | |
| 852 __ jcc(Assembler::negative, L); // NaN -> 0 | |
| 853 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong | |
| 854 __ mov64(c_rarg2, 0x8000000000000000); | |
| 855 __ mov64(rax, 0x7fffffffffffffff); | |
| 856 __ cmovq(Assembler::positive, c_rarg2, rax); | |
| 857 | |
| 858 __ bind(L); | |
| 859 __ movq(inout, c_rarg2); | |
| 860 | |
| 861 __ popq(c_rarg0); | |
| 862 __ popq(c_rarg1); | |
| 863 __ popq(c_rarg2); | |
| 864 __ popq(c_rarg3); | |
| 865 __ popq(rax); | |
| 866 | |
| 867 __ ret(0); | |
| 868 | |
| 869 return start; | |
| 870 } | |
| 871 | |
| 872 address generate_fp_mask(const char *stub_name, int64_t mask) { | |
| 873 StubCodeMark mark(this, "StubRoutines", stub_name); | |
| 874 | |
| 875 __ align(16); | |
| 876 address start = __ pc(); | |
| 877 | |
| 878 __ emit_data64( mask, relocInfo::none ); | |
| 879 __ emit_data64( mask, relocInfo::none ); | |
| 880 | |
| 881 return start; | |
| 882 } | |
| 883 | |
| 884 // The following routine generates a subroutine to throw an | |
| 885 // asynchronous UnknownError when an unsafe access gets a fault that | |
| 886 // could not be reasonably prevented by the programmer. (Example: | |
| 887 // SIGBUS/OBJERR.) | |
| 888 address generate_handler_for_unsafe_access() { | |
| 889 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access"); | |
| 890 address start = __ pc(); | |
| 891 | |
| 892 __ pushq(0); // hole for return address-to-be | |
| 893 __ pushaq(); // push registers | |
| 894 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord); | |
| 895 | |
| 896 __ subq(rsp, frame::arg_reg_save_area_bytes); | |
| 897 BLOCK_COMMENT("call handle_unsafe_access"); | |
| 898 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access))); | |
| 899 __ addq(rsp, frame::arg_reg_save_area_bytes); | |
| 900 | |
| 901 __ movq(next_pc, rax); // stuff next address | |
| 902 __ popaq(); | |
| 903 __ ret(0); // jump to next address | |
| 904 | |
| 905 return start; | |
| 906 } | |
| 907 | |
| 908 // Non-destructive plausibility checks for oops | |
| 909 // | |
| 910 // Arguments: | |
| 911 // all args on stack! | |
| 912 // | |
| 913 // Stack after saving c_rarg3: | |
| 914 // [tos + 0]: saved c_rarg3 | |
| 915 // [tos + 1]: saved c_rarg2 | |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
916 // [tos + 2]: saved r12 (several TemplateTable methods use it) |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
917 // [tos + 3]: saved flags |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
918 // [tos + 4]: return address |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
919 // * [tos + 5]: error message (char*) |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
920 // * [tos + 6]: object to verify (oop) |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
921 // * [tos + 7]: saved rax - saved by caller and bashed |
| 0 | 922 // * = popped on exit |
| 923 address generate_verify_oop() { | |
| 924 StubCodeMark mark(this, "StubRoutines", "verify_oop"); | |
| 925 address start = __ pc(); | |
| 926 | |
| 927 Label exit, error; | |
| 928 | |
| 929 __ pushfq(); | |
| 930 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr())); | |
| 931 | |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
932 __ pushq(r12); |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
933 |
| 0 | 934 // save c_rarg2 and c_rarg3 |
| 935 __ pushq(c_rarg2); | |
| 936 __ pushq(c_rarg3); | |
| 937 | |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
938 enum { |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
939 // After previous pushes. |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
940 oop_to_verify = 6 * wordSize, |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
941 saved_rax = 7 * wordSize, |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
942 |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
943 // Before the call to MacroAssembler::debug(), see below. |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
944 return_addr = 16 * wordSize, |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
945 error_msg = 17 * wordSize |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
946 }; |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
947 |
| 0 | 948 // get object |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
949 __ movq(rax, Address(rsp, oop_to_verify)); |
| 0 | 950 |
| 951 // make sure object is 'reasonable' | |
| 952 __ testq(rax, rax); | |
| 953 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK | |
| 954 // Check if the oop is in the right area of memory | |
| 955 __ movq(c_rarg2, rax); | |
| 956 __ movptr(c_rarg3, (int64_t) Universe::verify_oop_mask()); | |
| 957 __ andq(c_rarg2, c_rarg3); | |
| 958 __ movptr(c_rarg3, (int64_t) Universe::verify_oop_bits()); | |
| 959 __ cmpq(c_rarg2, c_rarg3); | |
| 960 __ jcc(Assembler::notZero, error); | |
| 961 | |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
962 // set r12 to heapbase for load_klass() |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
963 __ reinit_heapbase(); |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
964 |
| 0 | 965 // make sure klass is 'reasonable' |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
966 __ load_klass(rax, rax); // get klass |
| 0 | 967 __ testq(rax, rax); |
| 968 __ jcc(Assembler::zero, error); // if klass is NULL it is broken | |
| 969 // Check if the klass is in the right area of memory | |
| 970 __ movq(c_rarg2, rax); | |
| 971 __ movptr(c_rarg3, (int64_t) Universe::verify_klass_mask()); | |
| 972 __ andq(c_rarg2, c_rarg3); | |
| 973 __ movptr(c_rarg3, (int64_t) Universe::verify_klass_bits()); | |
| 974 __ cmpq(c_rarg2, c_rarg3); | |
| 975 __ jcc(Assembler::notZero, error); | |
| 976 | |
| 977 // make sure klass' klass is 'reasonable' | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
978 __ load_klass(rax, rax); |
| 0 | 979 __ testq(rax, rax); |
| 980 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken | |
| 981 // Check if the klass' klass is in the right area of memory | |
| 982 __ movptr(c_rarg3, (int64_t) Universe::verify_klass_mask()); | |
| 983 __ andq(rax, c_rarg3); | |
| 984 __ movptr(c_rarg3, (int64_t) Universe::verify_klass_bits()); | |
| 985 __ cmpq(rax, c_rarg3); | |
| 986 __ jcc(Assembler::notZero, error); | |
| 987 | |
| 988 // return if everything seems ok | |
| 989 __ bind(exit); | |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
990 __ movq(rax, Address(rsp, saved_rax)); // get saved rax back |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
991 __ popq(c_rarg3); // restore c_rarg3 |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
992 __ popq(c_rarg2); // restore c_rarg2 |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
993 __ popq(r12); // restore r12 |
| 0 | 994 __ popfq(); // restore flags |
| 995 __ ret(3 * wordSize); // pop caller saved stuff | |
| 996 | |
| 997 // handle errors | |
| 998 __ bind(error); | |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
999 __ movq(rax, Address(rsp, saved_rax)); // get saved rax back |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1000 __ popq(c_rarg3); // get saved c_rarg3 back |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1001 __ popq(c_rarg2); // get saved c_rarg2 back |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1002 __ popq(r12); // get saved r12 back |
| 0 | 1003 __ popfq(); // get saved flags off stack -- |
| 1004 // will be ignored | |
| 1005 | |
| 1006 __ pushaq(); // push registers | |
| 1007 // (rip is already | |
| 1008 // already pushed) | |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1009 // debug(char* msg, int64_t pc, int64_t regs[]) |
| 0 | 1010 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and |
| 1011 // pushed all the registers, so now the stack looks like: | |
| 1012 // [tos + 0] 16 saved registers | |
| 1013 // [tos + 16] return address | |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1014 // * [tos + 17] error message (char*) |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1015 // * [tos + 18] object to verify (oop) |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1016 // * [tos + 19] saved rax - saved by caller and bashed |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1017 // * = popped on exit |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1018 |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1019 __ movq(c_rarg0, Address(rsp, error_msg)); // pass address of error message |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1020 __ movq(c_rarg1, Address(rsp, return_addr)); // pass return address |
|
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1021 __ movq(c_rarg2, rsp); // pass address of regs on stack |
| 0 | 1022 __ movq(r12, rsp); // remember rsp |
| 1023 __ subq(rsp, frame::arg_reg_save_area_bytes);// windows | |
| 1024 __ andq(rsp, -16); // align stack as required by ABI | |
| 1025 BLOCK_COMMENT("call MacroAssembler::debug"); | |
| 1026 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug))); | |
| 1027 __ movq(rsp, r12); // restore rsp | |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1028 __ popaq(); // pop registers (includes r12) |
| 0 | 1029 __ ret(3 * wordSize); // pop caller saved stuff |
| 1030 | |
| 1031 return start; | |
| 1032 } | |
| 1033 | |
| 1034 static address disjoint_byte_copy_entry; | |
| 1035 static address disjoint_short_copy_entry; | |
| 1036 static address disjoint_int_copy_entry; | |
| 1037 static address disjoint_long_copy_entry; | |
| 1038 static address disjoint_oop_copy_entry; | |
| 1039 | |
| 1040 static address byte_copy_entry; | |
| 1041 static address short_copy_entry; | |
| 1042 static address int_copy_entry; | |
| 1043 static address long_copy_entry; | |
| 1044 static address oop_copy_entry; | |
| 1045 | |
| 1046 static address checkcast_copy_entry; | |
| 1047 | |
| 1048 // | |
| 1049 // Verify that a register contains clean 32-bits positive value | |
| 1050 // (high 32-bits are 0) so it could be used in 64-bits shifts. | |
| 1051 // | |
| 1052 // Input: | |
| 1053 // Rint - 32-bits value | |
| 1054 // Rtmp - scratch | |
| 1055 // | |
| 1056 void assert_clean_int(Register Rint, Register Rtmp) { | |
| 1057 #ifdef ASSERT | |
| 1058 Label L; | |
| 1059 assert_different_registers(Rtmp, Rint); | |
| 1060 __ movslq(Rtmp, Rint); | |
| 1061 __ cmpq(Rtmp, Rint); | |
|
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1062 __ jcc(Assembler::equal, L); |
| 0 | 1063 __ stop("high 32-bits of int value are not 0"); |
| 1064 __ bind(L); | |
| 1065 #endif | |
| 1066 } | |
| 1067 | |
| 1068 // Generate overlap test for array copy stubs | |
| 1069 // | |
| 1070 // Input: | |
| 1071 // c_rarg0 - from | |
| 1072 // c_rarg1 - to | |
| 1073 // c_rarg2 - element count | |
| 1074 // | |
| 1075 // Output: | |
| 1076 // rax - &from[element count - 1] | |
| 1077 // | |
| 1078 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) { | |
| 1079 assert(no_overlap_target != NULL, "must be generated"); | |
| 1080 array_overlap_test(no_overlap_target, NULL, sf); | |
| 1081 } | |
| 1082 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) { | |
| 1083 array_overlap_test(NULL, &L_no_overlap, sf); | |
| 1084 } | |
| 1085 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) { | |
| 1086 const Register from = c_rarg0; | |
| 1087 const Register to = c_rarg1; | |
| 1088 const Register count = c_rarg2; | |
| 1089 const Register end_from = rax; | |
| 1090 | |
| 1091 __ cmpq(to, from); | |
| 1092 __ leaq(end_from, Address(from, count, sf, 0)); | |
| 1093 if (NOLp == NULL) { | |
| 1094 ExternalAddress no_overlap(no_overlap_target); | |
| 1095 __ jump_cc(Assembler::belowEqual, no_overlap); | |
| 1096 __ cmpq(to, end_from); | |
| 1097 __ jump_cc(Assembler::aboveEqual, no_overlap); | |
| 1098 } else { | |
| 1099 __ jcc(Assembler::belowEqual, (*NOLp)); | |
| 1100 __ cmpq(to, end_from); | |
| 1101 __ jcc(Assembler::aboveEqual, (*NOLp)); | |
| 1102 } | |
| 1103 } | |
| 1104 | |
| 1105 // Shuffle first three arg regs on Windows into Linux/Solaris locations. | |
| 1106 // | |
| 1107 // Outputs: | |
| 1108 // rdi - rcx | |
| 1109 // rsi - rdx | |
| 1110 // rdx - r8 | |
| 1111 // rcx - r9 | |
| 1112 // | |
| 1113 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter | |
| 1114 // are non-volatile. r9 and r10 should not be used by the caller. | |
| 1115 // | |
| 1116 void setup_arg_regs(int nargs = 3) { | |
| 1117 const Register saved_rdi = r9; | |
| 1118 const Register saved_rsi = r10; | |
| 1119 assert(nargs == 3 || nargs == 4, "else fix"); | |
| 1120 #ifdef _WIN64 | |
| 1121 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9, | |
| 1122 "unexpected argument registers"); | |
| 1123 if (nargs >= 4) | |
| 1124 __ movq(rax, r9); // r9 is also saved_rdi | |
| 1125 __ movq(saved_rdi, rdi); | |
| 1126 __ movq(saved_rsi, rsi); | |
| 1127 __ movq(rdi, rcx); // c_rarg0 | |
| 1128 __ movq(rsi, rdx); // c_rarg1 | |
| 1129 __ movq(rdx, r8); // c_rarg2 | |
| 1130 if (nargs >= 4) | |
| 1131 __ movq(rcx, rax); // c_rarg3 (via rax) | |
| 1132 #else | |
| 1133 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx, | |
| 1134 "unexpected argument registers"); | |
| 1135 #endif | |
| 1136 } | |
| 1137 | |
| 1138 void restore_arg_regs() { | |
| 1139 const Register saved_rdi = r9; | |
| 1140 const Register saved_rsi = r10; | |
| 1141 #ifdef _WIN64 | |
| 1142 __ movq(rdi, saved_rdi); | |
| 1143 __ movq(rsi, saved_rsi); | |
| 1144 #endif | |
| 1145 } | |
| 1146 | |
| 1147 // Generate code for an array write pre barrier | |
| 1148 // | |
| 1149 // addr - starting address | |
| 1150 // count - element count | |
| 1151 // | |
| 1152 // Destroy no registers! | |
| 1153 // | |
| 1154 void gen_write_ref_array_pre_barrier(Register addr, Register count) { | |
| 1155 BarrierSet* bs = Universe::heap()->barrier_set(); | |
| 1156 switch (bs->kind()) { | |
| 1157 case BarrierSet::G1SATBCT: | |
| 1158 case BarrierSet::G1SATBCTLogging: | |
| 1159 { | |
| 1160 __ pushaq(); // push registers | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1161 if (count == c_rarg0) { |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1162 if (addr == c_rarg1) { |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1163 // exactly backwards!! |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1164 __ xchgq(c_rarg1, c_rarg0); |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1165 } else { |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1166 __ movq(c_rarg1, count); |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1167 __ movq(c_rarg0, addr); |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1168 } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1169 |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1170 } else { |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1171 __ movq(c_rarg0, addr); |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1172 __ movq(c_rarg1, count); |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1173 } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1174 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre))); |
| 0 | 1175 __ popaq(); |
| 1176 } | |
| 1177 break; | |
| 1178 case BarrierSet::CardTableModRef: | |
| 1179 case BarrierSet::CardTableExtension: | |
| 1180 case BarrierSet::ModRef: | |
| 1181 break; | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1182 default: |
| 0 | 1183 ShouldNotReachHere(); |
| 1184 | |
| 1185 } | |
| 1186 } | |
| 1187 | |
| 1188 // | |
| 1189 // Generate code for an array write post barrier | |
| 1190 // | |
| 1191 // Input: | |
| 1192 // start - register containing starting address of destination array | |
| 1193 // end - register containing ending address of destination array | |
| 1194 // scratch - scratch register | |
| 1195 // | |
| 1196 // The input registers are overwritten. | |
| 1197 // The ending address is inclusive. | |
| 1198 void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) { | |
| 1199 assert_different_registers(start, end, scratch); | |
| 1200 BarrierSet* bs = Universe::heap()->barrier_set(); | |
| 1201 switch (bs->kind()) { | |
| 1202 case BarrierSet::G1SATBCT: | |
| 1203 case BarrierSet::G1SATBCTLogging: | |
| 1204 | |
| 1205 { | |
| 1206 __ pushaq(); // push registers (overkill) | |
| 1207 // must compute element count unless barrier set interface is changed (other platforms supply count) | |
| 1208 assert_different_registers(start, end, scratch); | |
| 1209 __ leaq(scratch, Address(end, wordSize)); | |
| 1210 __ subq(scratch, start); | |
| 1211 __ shrq(scratch, LogBytesPerWord); | |
| 1212 __ movq(c_rarg0, start); | |
| 1213 __ movq(c_rarg1, scratch); | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1214 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post))); |
| 0 | 1215 __ popaq(); |
| 1216 } | |
| 1217 break; | |
| 1218 case BarrierSet::CardTableModRef: | |
| 1219 case BarrierSet::CardTableExtension: | |
| 1220 { | |
| 1221 CardTableModRefBS* ct = (CardTableModRefBS*)bs; | |
| 1222 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); | |
| 1223 | |
| 1224 Label L_loop; | |
| 1225 | |
| 1226 __ shrq(start, CardTableModRefBS::card_shift); | |
| 1227 __ shrq(end, CardTableModRefBS::card_shift); | |
| 1228 __ subq(end, start); // number of bytes to copy | |
| 1229 | |
|
249
910a4cb98e9e
6717457: Internal Error (src/share/vm/code/relocInfo.hpp:1089)
never
parents:
196
diff
changeset
|
1230 intptr_t disp = (intptr_t) ct->byte_map_base; |
|
910a4cb98e9e
6717457: Internal Error (src/share/vm/code/relocInfo.hpp:1089)
never
parents:
196
diff
changeset
|
1231 if (__ is_simm32(disp)) { |
|
910a4cb98e9e
6717457: Internal Error (src/share/vm/code/relocInfo.hpp:1089)
never
parents:
196
diff
changeset
|
1232 Address cardtable(noreg, noreg, Address::no_scale, disp); |
|
910a4cb98e9e
6717457: Internal Error (src/share/vm/code/relocInfo.hpp:1089)
never
parents:
196
diff
changeset
|
1233 __ lea(scratch, cardtable); |
|
910a4cb98e9e
6717457: Internal Error (src/share/vm/code/relocInfo.hpp:1089)
never
parents:
196
diff
changeset
|
1234 } else { |
|
910a4cb98e9e
6717457: Internal Error (src/share/vm/code/relocInfo.hpp:1089)
never
parents:
196
diff
changeset
|
1235 ExternalAddress cardtable((address)disp); |
|
910a4cb98e9e
6717457: Internal Error (src/share/vm/code/relocInfo.hpp:1089)
never
parents:
196
diff
changeset
|
1236 __ lea(scratch, cardtable); |
|
910a4cb98e9e
6717457: Internal Error (src/share/vm/code/relocInfo.hpp:1089)
never
parents:
196
diff
changeset
|
1237 } |
|
910a4cb98e9e
6717457: Internal Error (src/share/vm/code/relocInfo.hpp:1089)
never
parents:
196
diff
changeset
|
1238 |
| 0 | 1239 const Register count = end; // 'end' register contains bytes count now |
| 1240 __ addq(start, scratch); | |
| 1241 __ BIND(L_loop); | |
| 1242 __ movb(Address(start, count, Address::times_1), 0); | |
| 1243 __ decrementq(count); | |
| 1244 __ jcc(Assembler::greaterEqual, L_loop); | |
| 1245 } | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1246 break; |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1247 default: |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1248 ShouldNotReachHere(); |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1249 |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1250 } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1251 } |
| 0 | 1252 |
| 1253 // Copy big chunks forward | |
| 1254 // | |
| 1255 // Inputs: | |
| 1256 // end_from - source arrays end address | |
| 1257 // end_to - destination array end address | |
| 1258 // qword_count - 64-bits element count, negative | |
| 1259 // to - scratch | |
| 1260 // L_copy_32_bytes - entry label | |
| 1261 // L_copy_8_bytes - exit label | |
| 1262 // | |
| 1263 void copy_32_bytes_forward(Register end_from, Register end_to, | |
| 1264 Register qword_count, Register to, | |
| 1265 Label& L_copy_32_bytes, Label& L_copy_8_bytes) { | |
| 1266 DEBUG_ONLY(__ stop("enter at entry label, not here")); | |
| 1267 Label L_loop; | |
| 1268 __ align(16); | |
| 1269 __ BIND(L_loop); | |
| 1270 __ movq(to, Address(end_from, qword_count, Address::times_8, -24)); | |
| 1271 __ movq(Address(end_to, qword_count, Address::times_8, -24), to); | |
| 1272 __ movq(to, Address(end_from, qword_count, Address::times_8, -16)); | |
| 1273 __ movq(Address(end_to, qword_count, Address::times_8, -16), to); | |
| 1274 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8)); | |
| 1275 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to); | |
| 1276 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0)); | |
| 1277 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to); | |
| 1278 __ BIND(L_copy_32_bytes); | |
| 1279 __ addq(qword_count, 4); | |
| 1280 __ jcc(Assembler::lessEqual, L_loop); | |
| 1281 __ subq(qword_count, 4); | |
| 1282 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords | |
| 1283 } | |
| 1284 | |
| 1285 | |
| 1286 // Copy big chunks backward | |
| 1287 // | |
| 1288 // Inputs: | |
| 1289 // from - source arrays address | |
| 1290 // dest - destination array address | |
| 1291 // qword_count - 64-bits element count | |
| 1292 // to - scratch | |
| 1293 // L_copy_32_bytes - entry label | |
| 1294 // L_copy_8_bytes - exit label | |
| 1295 // | |
| 1296 void copy_32_bytes_backward(Register from, Register dest, | |
| 1297 Register qword_count, Register to, | |
| 1298 Label& L_copy_32_bytes, Label& L_copy_8_bytes) { | |
| 1299 DEBUG_ONLY(__ stop("enter at entry label, not here")); | |
| 1300 Label L_loop; | |
| 1301 __ align(16); | |
| 1302 __ BIND(L_loop); | |
| 1303 __ movq(to, Address(from, qword_count, Address::times_8, 24)); | |
| 1304 __ movq(Address(dest, qword_count, Address::times_8, 24), to); | |
| 1305 __ movq(to, Address(from, qword_count, Address::times_8, 16)); | |
| 1306 __ movq(Address(dest, qword_count, Address::times_8, 16), to); | |
| 1307 __ movq(to, Address(from, qword_count, Address::times_8, 8)); | |
| 1308 __ movq(Address(dest, qword_count, Address::times_8, 8), to); | |
| 1309 __ movq(to, Address(from, qword_count, Address::times_8, 0)); | |
| 1310 __ movq(Address(dest, qword_count, Address::times_8, 0), to); | |
| 1311 __ BIND(L_copy_32_bytes); | |
| 1312 __ subq(qword_count, 4); | |
| 1313 __ jcc(Assembler::greaterEqual, L_loop); | |
| 1314 __ addq(qword_count, 4); | |
| 1315 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords | |
| 1316 } | |
| 1317 | |
| 1318 | |
| 1319 // Arguments: | |
| 1320 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary | |
| 1321 // ignored | |
| 1322 // name - stub name string | |
| 1323 // | |
| 1324 // Inputs: | |
| 1325 // c_rarg0 - source array address | |
| 1326 // c_rarg1 - destination array address | |
| 1327 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1328 // | |
| 1329 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries, | |
| 1330 // we let the hardware handle it. The one to eight bytes within words, | |
| 1331 // dwords or qwords that span cache line boundaries will still be loaded | |
| 1332 // and stored atomically. | |
| 1333 // | |
| 1334 // Side Effects: | |
| 1335 // disjoint_byte_copy_entry is set to the no-overlap entry point | |
| 1336 // used by generate_conjoint_byte_copy(). | |
| 1337 // | |
| 1338 address generate_disjoint_byte_copy(bool aligned, const char *name) { | |
| 1339 __ align(CodeEntryAlignment); | |
| 1340 StubCodeMark mark(this, "StubRoutines", name); | |
| 1341 address start = __ pc(); | |
| 1342 | |
| 1343 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes; | |
| 1344 Label L_copy_byte, L_exit; | |
| 1345 const Register from = rdi; // source array address | |
| 1346 const Register to = rsi; // destination array address | |
| 1347 const Register count = rdx; // elements count | |
| 1348 const Register byte_count = rcx; | |
| 1349 const Register qword_count = count; | |
| 1350 const Register end_from = from; // source array end address | |
| 1351 const Register end_to = to; // destination array end address | |
| 1352 // End pointers are inclusive, and if count is not zero they point | |
| 1353 // to the last unit copied: end_to[0] := end_from[0] | |
| 1354 | |
| 1355 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1356 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1357 | |
| 1358 disjoint_byte_copy_entry = __ pc(); | |
| 1359 BLOCK_COMMENT("Entry:"); | |
| 1360 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1361 | |
| 1362 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1363 // r9 and r10 may be used to save non-volatile registers | |
| 1364 | |
| 1365 // 'from', 'to' and 'count' are now valid | |
| 1366 __ movq(byte_count, count); | |
| 1367 __ shrq(count, 3); // count => qword_count | |
| 1368 | |
| 1369 // Copy from low to high addresses. Use 'to' as scratch. | |
| 1370 __ leaq(end_from, Address(from, qword_count, Address::times_8, -8)); | |
| 1371 __ leaq(end_to, Address(to, qword_count, Address::times_8, -8)); | |
| 1372 __ negq(qword_count); // make the count negative | |
| 1373 __ jmp(L_copy_32_bytes); | |
| 1374 | |
| 1375 // Copy trailing qwords | |
| 1376 __ BIND(L_copy_8_bytes); | |
| 1377 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); | |
| 1378 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); | |
| 1379 __ incrementq(qword_count); | |
| 1380 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1381 | |
| 1382 // Check for and copy trailing dword | |
| 1383 __ BIND(L_copy_4_bytes); | |
| 1384 __ testq(byte_count, 4); | |
| 1385 __ jccb(Assembler::zero, L_copy_2_bytes); | |
| 1386 __ movl(rax, Address(end_from, 8)); | |
| 1387 __ movl(Address(end_to, 8), rax); | |
| 1388 | |
| 1389 __ addq(end_from, 4); | |
| 1390 __ addq(end_to, 4); | |
| 1391 | |
| 1392 // Check for and copy trailing word | |
| 1393 __ BIND(L_copy_2_bytes); | |
| 1394 __ testq(byte_count, 2); | |
| 1395 __ jccb(Assembler::zero, L_copy_byte); | |
| 1396 __ movw(rax, Address(end_from, 8)); | |
| 1397 __ movw(Address(end_to, 8), rax); | |
| 1398 | |
| 1399 __ addq(end_from, 2); | |
| 1400 __ addq(end_to, 2); | |
| 1401 | |
| 1402 // Check for and copy trailing byte | |
| 1403 __ BIND(L_copy_byte); | |
| 1404 __ testq(byte_count, 1); | |
| 1405 __ jccb(Assembler::zero, L_exit); | |
| 1406 __ movb(rax, Address(end_from, 8)); | |
| 1407 __ movb(Address(end_to, 8), rax); | |
| 1408 | |
| 1409 __ BIND(L_exit); | |
| 1410 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); | |
| 1411 restore_arg_regs(); | |
| 1412 __ xorq(rax, rax); // return 0 | |
| 1413 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1414 __ ret(0); | |
| 1415 | |
| 1416 // Copy in 32-bytes chunks | |
| 1417 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1418 __ jmp(L_copy_4_bytes); | |
| 1419 | |
| 1420 return start; | |
| 1421 } | |
| 1422 | |
| 1423 // Arguments: | |
| 1424 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary | |
| 1425 // ignored | |
| 1426 // name - stub name string | |
| 1427 // | |
| 1428 // Inputs: | |
| 1429 // c_rarg0 - source array address | |
| 1430 // c_rarg1 - destination array address | |
| 1431 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1432 // | |
| 1433 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries, | |
| 1434 // we let the hardware handle it. The one to eight bytes within words, | |
| 1435 // dwords or qwords that span cache line boundaries will still be loaded | |
| 1436 // and stored atomically. | |
| 1437 // | |
| 1438 address generate_conjoint_byte_copy(bool aligned, const char *name) { | |
| 1439 __ align(CodeEntryAlignment); | |
| 1440 StubCodeMark mark(this, "StubRoutines", name); | |
| 1441 address start = __ pc(); | |
| 1442 | |
| 1443 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes; | |
| 1444 const Register from = rdi; // source array address | |
| 1445 const Register to = rsi; // destination array address | |
| 1446 const Register count = rdx; // elements count | |
| 1447 const Register byte_count = rcx; | |
| 1448 const Register qword_count = count; | |
| 1449 | |
| 1450 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1451 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1452 | |
| 1453 byte_copy_entry = __ pc(); | |
| 1454 BLOCK_COMMENT("Entry:"); | |
| 1455 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1456 | |
| 1457 array_overlap_test(disjoint_byte_copy_entry, Address::times_1); | |
| 1458 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1459 // r9 and r10 may be used to save non-volatile registers | |
| 1460 | |
| 1461 // 'from', 'to' and 'count' are now valid | |
| 1462 __ movq(byte_count, count); | |
| 1463 __ shrq(count, 3); // count => qword_count | |
| 1464 | |
| 1465 // Copy from high to low addresses. | |
| 1466 | |
| 1467 // Check for and copy trailing byte | |
| 1468 __ testq(byte_count, 1); | |
| 1469 __ jcc(Assembler::zero, L_copy_2_bytes); | |
| 1470 __ movb(rax, Address(from, byte_count, Address::times_1, -1)); | |
| 1471 __ movb(Address(to, byte_count, Address::times_1, -1), rax); | |
| 1472 __ decrementq(byte_count); // Adjust for possible trailing word | |
| 1473 | |
| 1474 // Check for and copy trailing word | |
| 1475 __ BIND(L_copy_2_bytes); | |
| 1476 __ testq(byte_count, 2); | |
| 1477 __ jcc(Assembler::zero, L_copy_4_bytes); | |
| 1478 __ movw(rax, Address(from, byte_count, Address::times_1, -2)); | |
| 1479 __ movw(Address(to, byte_count, Address::times_1, -2), rax); | |
| 1480 | |
| 1481 // Check for and copy trailing dword | |
| 1482 __ BIND(L_copy_4_bytes); | |
| 1483 __ testq(byte_count, 4); | |
| 1484 __ jcc(Assembler::zero, L_copy_32_bytes); | |
| 1485 __ movl(rax, Address(from, qword_count, Address::times_8)); | |
| 1486 __ movl(Address(to, qword_count, Address::times_8), rax); | |
| 1487 __ jmp(L_copy_32_bytes); | |
| 1488 | |
| 1489 // Copy trailing qwords | |
| 1490 __ BIND(L_copy_8_bytes); | |
| 1491 __ movq(rax, Address(from, qword_count, Address::times_8, -8)); | |
| 1492 __ movq(Address(to, qword_count, Address::times_8, -8), rax); | |
| 1493 __ decrementq(qword_count); | |
| 1494 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1495 | |
| 1496 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); | |
| 1497 restore_arg_regs(); | |
| 1498 __ xorq(rax, rax); // return 0 | |
| 1499 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1500 __ ret(0); | |
| 1501 | |
| 1502 // Copy in 32-bytes chunks | |
| 1503 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1504 | |
| 1505 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); | |
| 1506 restore_arg_regs(); | |
| 1507 __ xorq(rax, rax); // return 0 | |
| 1508 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1509 __ ret(0); | |
| 1510 | |
| 1511 return start; | |
| 1512 } | |
| 1513 | |
| 1514 // Arguments: | |
| 1515 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary | |
| 1516 // ignored | |
| 1517 // name - stub name string | |
| 1518 // | |
| 1519 // Inputs: | |
| 1520 // c_rarg0 - source array address | |
| 1521 // c_rarg1 - destination array address | |
| 1522 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1523 // | |
| 1524 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we | |
| 1525 // let the hardware handle it. The two or four words within dwords | |
| 1526 // or qwords that span cache line boundaries will still be loaded | |
| 1527 // and stored atomically. | |
| 1528 // | |
| 1529 // Side Effects: | |
| 1530 // disjoint_short_copy_entry is set to the no-overlap entry point | |
| 1531 // used by generate_conjoint_short_copy(). | |
| 1532 // | |
| 1533 address generate_disjoint_short_copy(bool aligned, const char *name) { | |
| 1534 __ align(CodeEntryAlignment); | |
| 1535 StubCodeMark mark(this, "StubRoutines", name); | |
| 1536 address start = __ pc(); | |
| 1537 | |
| 1538 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit; | |
| 1539 const Register from = rdi; // source array address | |
| 1540 const Register to = rsi; // destination array address | |
| 1541 const Register count = rdx; // elements count | |
| 1542 const Register word_count = rcx; | |
| 1543 const Register qword_count = count; | |
| 1544 const Register end_from = from; // source array end address | |
| 1545 const Register end_to = to; // destination array end address | |
| 1546 // End pointers are inclusive, and if count is not zero they point | |
| 1547 // to the last unit copied: end_to[0] := end_from[0] | |
| 1548 | |
| 1549 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1550 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1551 | |
| 1552 disjoint_short_copy_entry = __ pc(); | |
| 1553 BLOCK_COMMENT("Entry:"); | |
| 1554 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1555 | |
| 1556 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1557 // r9 and r10 may be used to save non-volatile registers | |
| 1558 | |
| 1559 // 'from', 'to' and 'count' are now valid | |
| 1560 __ movq(word_count, count); | |
| 1561 __ shrq(count, 2); // count => qword_count | |
| 1562 | |
| 1563 // Copy from low to high addresses. Use 'to' as scratch. | |
| 1564 __ leaq(end_from, Address(from, qword_count, Address::times_8, -8)); | |
| 1565 __ leaq(end_to, Address(to, qword_count, Address::times_8, -8)); | |
| 1566 __ negq(qword_count); | |
| 1567 __ jmp(L_copy_32_bytes); | |
| 1568 | |
| 1569 // Copy trailing qwords | |
| 1570 __ BIND(L_copy_8_bytes); | |
| 1571 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); | |
| 1572 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); | |
| 1573 __ incrementq(qword_count); | |
| 1574 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1575 | |
| 1576 // Original 'dest' is trashed, so we can't use it as a | |
| 1577 // base register for a possible trailing word copy | |
| 1578 | |
| 1579 // Check for and copy trailing dword | |
| 1580 __ BIND(L_copy_4_bytes); | |
| 1581 __ testq(word_count, 2); | |
| 1582 __ jccb(Assembler::zero, L_copy_2_bytes); | |
| 1583 __ movl(rax, Address(end_from, 8)); | |
| 1584 __ movl(Address(end_to, 8), rax); | |
| 1585 | |
| 1586 __ addq(end_from, 4); | |
| 1587 __ addq(end_to, 4); | |
| 1588 | |
| 1589 // Check for and copy trailing word | |
| 1590 __ BIND(L_copy_2_bytes); | |
| 1591 __ testq(word_count, 1); | |
| 1592 __ jccb(Assembler::zero, L_exit); | |
| 1593 __ movw(rax, Address(end_from, 8)); | |
| 1594 __ movw(Address(end_to, 8), rax); | |
| 1595 | |
| 1596 __ BIND(L_exit); | |
| 1597 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); | |
| 1598 restore_arg_regs(); | |
| 1599 __ xorq(rax, rax); // return 0 | |
| 1600 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1601 __ ret(0); | |
| 1602 | |
| 1603 // Copy in 32-bytes chunks | |
| 1604 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1605 __ jmp(L_copy_4_bytes); | |
| 1606 | |
| 1607 return start; | |
| 1608 } | |
| 1609 | |
| 1610 // Arguments: | |
| 1611 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary | |
| 1612 // ignored | |
| 1613 // name - stub name string | |
| 1614 // | |
| 1615 // Inputs: | |
| 1616 // c_rarg0 - source array address | |
| 1617 // c_rarg1 - destination array address | |
| 1618 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1619 // | |
| 1620 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we | |
| 1621 // let the hardware handle it. The two or four words within dwords | |
| 1622 // or qwords that span cache line boundaries will still be loaded | |
| 1623 // and stored atomically. | |
| 1624 // | |
| 1625 address generate_conjoint_short_copy(bool aligned, const char *name) { | |
| 1626 __ align(CodeEntryAlignment); | |
| 1627 StubCodeMark mark(this, "StubRoutines", name); | |
| 1628 address start = __ pc(); | |
| 1629 | |
| 1630 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes; | |
| 1631 const Register from = rdi; // source array address | |
| 1632 const Register to = rsi; // destination array address | |
| 1633 const Register count = rdx; // elements count | |
| 1634 const Register word_count = rcx; | |
| 1635 const Register qword_count = count; | |
| 1636 | |
| 1637 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1638 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1639 | |
| 1640 short_copy_entry = __ pc(); | |
| 1641 BLOCK_COMMENT("Entry:"); | |
| 1642 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1643 | |
| 1644 array_overlap_test(disjoint_short_copy_entry, Address::times_2); | |
| 1645 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1646 // r9 and r10 may be used to save non-volatile registers | |
| 1647 | |
| 1648 // 'from', 'to' and 'count' are now valid | |
| 1649 __ movq(word_count, count); | |
| 1650 __ shrq(count, 2); // count => qword_count | |
| 1651 | |
| 1652 // Copy from high to low addresses. Use 'to' as scratch. | |
| 1653 | |
| 1654 // Check for and copy trailing word | |
| 1655 __ testq(word_count, 1); | |
| 1656 __ jccb(Assembler::zero, L_copy_4_bytes); | |
| 1657 __ movw(rax, Address(from, word_count, Address::times_2, -2)); | |
| 1658 __ movw(Address(to, word_count, Address::times_2, -2), rax); | |
| 1659 | |
| 1660 // Check for and copy trailing dword | |
| 1661 __ BIND(L_copy_4_bytes); | |
| 1662 __ testq(word_count, 2); | |
| 1663 __ jcc(Assembler::zero, L_copy_32_bytes); | |
| 1664 __ movl(rax, Address(from, qword_count, Address::times_8)); | |
| 1665 __ movl(Address(to, qword_count, Address::times_8), rax); | |
| 1666 __ jmp(L_copy_32_bytes); | |
| 1667 | |
| 1668 // Copy trailing qwords | |
| 1669 __ BIND(L_copy_8_bytes); | |
| 1670 __ movq(rax, Address(from, qword_count, Address::times_8, -8)); | |
| 1671 __ movq(Address(to, qword_count, Address::times_8, -8), rax); | |
| 1672 __ decrementq(qword_count); | |
| 1673 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1674 | |
| 1675 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); | |
| 1676 restore_arg_regs(); | |
| 1677 __ xorq(rax, rax); // return 0 | |
| 1678 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1679 __ ret(0); | |
| 1680 | |
| 1681 // Copy in 32-bytes chunks | |
| 1682 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1683 | |
| 1684 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); | |
| 1685 restore_arg_regs(); | |
| 1686 __ xorq(rax, rax); // return 0 | |
| 1687 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1688 __ ret(0); | |
| 1689 | |
| 1690 return start; | |
| 1691 } | |
| 1692 | |
| 1693 // Arguments: | |
| 1694 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary | |
| 1695 // ignored | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1696 // is_oop - true => oop array, so generate store check code |
| 0 | 1697 // name - stub name string |
| 1698 // | |
| 1699 // Inputs: | |
| 1700 // c_rarg0 - source array address | |
| 1701 // c_rarg1 - destination array address | |
| 1702 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1703 // | |
| 1704 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let | |
| 1705 // the hardware handle it. The two dwords within qwords that span | |
| 1706 // cache line boundaries will still be loaded and stored atomicly. | |
| 1707 // | |
| 1708 // Side Effects: | |
| 1709 // disjoint_int_copy_entry is set to the no-overlap entry point | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1710 // used by generate_conjoint_int_oop_copy(). |
| 0 | 1711 // |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1712 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) { |
| 0 | 1713 __ align(CodeEntryAlignment); |
| 1714 StubCodeMark mark(this, "StubRoutines", name); | |
| 1715 address start = __ pc(); | |
| 1716 | |
| 1717 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit; | |
| 1718 const Register from = rdi; // source array address | |
| 1719 const Register to = rsi; // destination array address | |
| 1720 const Register count = rdx; // elements count | |
| 1721 const Register dword_count = rcx; | |
| 1722 const Register qword_count = count; | |
| 1723 const Register end_from = from; // source array end address | |
| 1724 const Register end_to = to; // destination array end address | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1725 const Register saved_to = r11; // saved destination array address |
| 0 | 1726 // End pointers are inclusive, and if count is not zero they point |
| 1727 // to the last unit copied: end_to[0] := end_from[0] | |
| 1728 | |
| 1729 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1730 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1731 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1732 (is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry) = __ pc(); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1733 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1734 if (is_oop) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1735 // no registers are destroyed by this call |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1736 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1737 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1738 |
| 0 | 1739 BLOCK_COMMENT("Entry:"); |
| 1740 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1741 | |
| 1742 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1743 // r9 and r10 may be used to save non-volatile registers | |
| 1744 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1745 if (is_oop) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1746 __ movq(saved_to, to); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1747 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1748 |
| 0 | 1749 // 'from', 'to' and 'count' are now valid |
| 1750 __ movq(dword_count, count); | |
| 1751 __ shrq(count, 1); // count => qword_count | |
| 1752 | |
| 1753 // Copy from low to high addresses. Use 'to' as scratch. | |
| 1754 __ leaq(end_from, Address(from, qword_count, Address::times_8, -8)); | |
| 1755 __ leaq(end_to, Address(to, qword_count, Address::times_8, -8)); | |
| 1756 __ negq(qword_count); | |
| 1757 __ jmp(L_copy_32_bytes); | |
| 1758 | |
| 1759 // Copy trailing qwords | |
| 1760 __ BIND(L_copy_8_bytes); | |
| 1761 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); | |
| 1762 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); | |
| 1763 __ incrementq(qword_count); | |
| 1764 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1765 | |
| 1766 // Check for and copy trailing dword | |
| 1767 __ BIND(L_copy_4_bytes); | |
| 1768 __ testq(dword_count, 1); // Only byte test since the value is 0 or 1 | |
| 1769 __ jccb(Assembler::zero, L_exit); | |
| 1770 __ movl(rax, Address(end_from, 8)); | |
| 1771 __ movl(Address(end_to, 8), rax); | |
| 1772 | |
| 1773 __ BIND(L_exit); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1774 if (is_oop) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1775 __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4)); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1776 gen_write_ref_array_post_barrier(saved_to, end_to, rax); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1777 } |
| 0 | 1778 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); |
| 1779 restore_arg_regs(); | |
| 1780 __ xorq(rax, rax); // return 0 | |
| 1781 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1782 __ ret(0); | |
| 1783 | |
| 1784 // Copy 32-bytes chunks | |
| 1785 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1786 __ jmp(L_copy_4_bytes); | |
| 1787 | |
| 1788 return start; | |
| 1789 } | |
| 1790 | |
| 1791 // Arguments: | |
| 1792 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary | |
| 1793 // ignored | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1794 // is_oop - true => oop array, so generate store check code |
| 0 | 1795 // name - stub name string |
| 1796 // | |
| 1797 // Inputs: | |
| 1798 // c_rarg0 - source array address | |
| 1799 // c_rarg1 - destination array address | |
| 1800 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1801 // | |
| 1802 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let | |
| 1803 // the hardware handle it. The two dwords within qwords that span | |
| 1804 // cache line boundaries will still be loaded and stored atomicly. | |
| 1805 // | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1806 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) { |
| 0 | 1807 __ align(CodeEntryAlignment); |
| 1808 StubCodeMark mark(this, "StubRoutines", name); | |
| 1809 address start = __ pc(); | |
| 1810 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1811 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit; |
| 0 | 1812 const Register from = rdi; // source array address |
| 1813 const Register to = rsi; // destination array address | |
| 1814 const Register count = rdx; // elements count | |
| 1815 const Register dword_count = rcx; | |
| 1816 const Register qword_count = count; | |
| 1817 | |
| 1818 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1819 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1820 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1821 if (is_oop) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1822 // no registers are destroyed by this call |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1823 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1824 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1825 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1826 (is_oop ? oop_copy_entry : int_copy_entry) = __ pc(); |
| 0 | 1827 BLOCK_COMMENT("Entry:"); |
| 1828 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1829 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1830 array_overlap_test(is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry, |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1831 Address::times_4); |
| 0 | 1832 setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
| 1833 // r9 and r10 may be used to save non-volatile registers | |
| 1834 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1835 assert_clean_int(count, rax); // Make sure 'count' is clean int. |
| 0 | 1836 // 'from', 'to' and 'count' are now valid |
| 1837 __ movq(dword_count, count); | |
| 1838 __ shrq(count, 1); // count => qword_count | |
| 1839 | |
| 1840 // Copy from high to low addresses. Use 'to' as scratch. | |
| 1841 | |
| 1842 // Check for and copy trailing dword | |
| 1843 __ testq(dword_count, 1); | |
| 1844 __ jcc(Assembler::zero, L_copy_32_bytes); | |
| 1845 __ movl(rax, Address(from, dword_count, Address::times_4, -4)); | |
| 1846 __ movl(Address(to, dword_count, Address::times_4, -4), rax); | |
| 1847 __ jmp(L_copy_32_bytes); | |
| 1848 | |
| 1849 // Copy trailing qwords | |
| 1850 __ BIND(L_copy_8_bytes); | |
| 1851 __ movq(rax, Address(from, qword_count, Address::times_8, -8)); | |
| 1852 __ movq(Address(to, qword_count, Address::times_8, -8), rax); | |
| 1853 __ decrementq(qword_count); | |
| 1854 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1855 | |
| 1856 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1857 if (is_oop) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1858 __ jmp(L_exit); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1859 } |
| 0 | 1860 restore_arg_regs(); |
| 1861 __ xorq(rax, rax); // return 0 | |
| 1862 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1863 __ ret(0); | |
| 1864 | |
| 1865 // Copy in 32-bytes chunks | |
| 1866 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1867 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1868 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1869 __ bind(L_exit); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1870 if (is_oop) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1871 Register end_to = rdx; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1872 __ leaq(end_to, Address(to, dword_count, Address::times_4, -4)); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1873 gen_write_ref_array_post_barrier(to, end_to, rax); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1874 } |
| 0 | 1875 restore_arg_regs(); |
| 1876 __ xorq(rax, rax); // return 0 | |
| 1877 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1878 __ ret(0); | |
| 1879 | |
| 1880 return start; | |
| 1881 } | |
| 1882 | |
| 1883 // Arguments: | |
| 1884 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes | |
| 1885 // ignored | |
| 1886 // is_oop - true => oop array, so generate store check code | |
| 1887 // name - stub name string | |
| 1888 // | |
| 1889 // Inputs: | |
| 1890 // c_rarg0 - source array address | |
| 1891 // c_rarg1 - destination array address | |
| 1892 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1893 // | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1894 // Side Effects: |
| 0 | 1895 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the |
| 1896 // no-overlap entry point used by generate_conjoint_long_oop_copy(). | |
| 1897 // | |
| 1898 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) { | |
| 1899 __ align(CodeEntryAlignment); | |
| 1900 StubCodeMark mark(this, "StubRoutines", name); | |
| 1901 address start = __ pc(); | |
| 1902 | |
| 1903 Label L_copy_32_bytes, L_copy_8_bytes, L_exit; | |
| 1904 const Register from = rdi; // source array address | |
| 1905 const Register to = rsi; // destination array address | |
| 1906 const Register qword_count = rdx; // elements count | |
| 1907 const Register end_from = from; // source array end address | |
| 1908 const Register end_to = rcx; // destination array end address | |
| 1909 const Register saved_to = to; | |
| 1910 // End pointers are inclusive, and if count is not zero they point | |
| 1911 // to the last unit copied: end_to[0] := end_from[0] | |
| 1912 | |
| 1913 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1914 // Save no-overlap entry point for generate_conjoint_long_oop_copy() | |
| 1915 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1916 | |
| 1917 if (is_oop) { | |
| 1918 disjoint_oop_copy_entry = __ pc(); | |
| 1919 // no registers are destroyed by this call | |
| 1920 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2); | |
| 1921 } else { | |
| 1922 disjoint_long_copy_entry = __ pc(); | |
| 1923 } | |
| 1924 BLOCK_COMMENT("Entry:"); | |
| 1925 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1926 | |
| 1927 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1928 // r9 and r10 may be used to save non-volatile registers | |
| 1929 | |
| 1930 // 'from', 'to' and 'qword_count' are now valid | |
| 1931 | |
| 1932 // Copy from low to high addresses. Use 'to' as scratch. | |
| 1933 __ leaq(end_from, Address(from, qword_count, Address::times_8, -8)); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1934 __ leaq(end_to, Address(to, qword_count, Address::times_8, -8)); |
| 0 | 1935 __ negq(qword_count); |
| 1936 __ jmp(L_copy_32_bytes); | |
| 1937 | |
| 1938 // Copy trailing qwords | |
| 1939 __ BIND(L_copy_8_bytes); | |
| 1940 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); | |
| 1941 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); | |
| 1942 __ incrementq(qword_count); | |
| 1943 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1944 | |
| 1945 if (is_oop) { | |
| 1946 __ jmp(L_exit); | |
| 1947 } else { | |
| 1948 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); | |
| 1949 restore_arg_regs(); | |
| 1950 __ xorq(rax, rax); // return 0 | |
| 1951 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1952 __ ret(0); | |
| 1953 } | |
| 1954 | |
| 1955 // Copy 64-byte chunks | |
| 1956 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1957 | |
| 1958 if (is_oop) { | |
| 1959 __ BIND(L_exit); | |
| 1960 gen_write_ref_array_post_barrier(saved_to, end_to, rax); | |
| 1961 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); | |
| 1962 } else { | |
| 1963 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); | |
| 1964 } | |
| 1965 restore_arg_regs(); | |
| 1966 __ xorq(rax, rax); // return 0 | |
| 1967 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1968 __ ret(0); | |
| 1969 | |
| 1970 return start; | |
| 1971 } | |
| 1972 | |
| 1973 // Arguments: | |
| 1974 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes | |
| 1975 // ignored | |
| 1976 // is_oop - true => oop array, so generate store check code | |
| 1977 // name - stub name string | |
| 1978 // | |
| 1979 // Inputs: | |
| 1980 // c_rarg0 - source array address | |
| 1981 // c_rarg1 - destination array address | |
| 1982 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1983 // | |
| 1984 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) { | |
| 1985 __ align(CodeEntryAlignment); | |
| 1986 StubCodeMark mark(this, "StubRoutines", name); | |
| 1987 address start = __ pc(); | |
| 1988 | |
| 1989 Label L_copy_32_bytes, L_copy_8_bytes, L_exit; | |
| 1990 const Register from = rdi; // source array address | |
| 1991 const Register to = rsi; // destination array address | |
| 1992 const Register qword_count = rdx; // elements count | |
| 1993 const Register saved_count = rcx; | |
| 1994 | |
| 1995 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1996 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1997 | |
| 1998 address disjoint_copy_entry = NULL; | |
| 1999 if (is_oop) { | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2000 assert(!UseCompressedOops, "shouldn't be called for compressed oops"); |
| 0 | 2001 disjoint_copy_entry = disjoint_oop_copy_entry; |
| 2002 oop_copy_entry = __ pc(); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2003 array_overlap_test(disjoint_oop_copy_entry, Address::times_8); |
| 0 | 2004 } else { |
| 2005 disjoint_copy_entry = disjoint_long_copy_entry; | |
| 2006 long_copy_entry = __ pc(); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2007 array_overlap_test(disjoint_long_copy_entry, Address::times_8); |
| 0 | 2008 } |
| 2009 BLOCK_COMMENT("Entry:"); | |
| 2010 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 2011 | |
| 2012 array_overlap_test(disjoint_copy_entry, Address::times_8); | |
| 2013 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 2014 // r9 and r10 may be used to save non-volatile registers | |
| 2015 | |
| 2016 // 'from', 'to' and 'qword_count' are now valid | |
| 2017 | |
| 2018 if (is_oop) { | |
| 2019 // Save to and count for store barrier | |
| 2020 __ movq(saved_count, qword_count); | |
| 2021 // No registers are destroyed by this call | |
| 2022 gen_write_ref_array_pre_barrier(to, saved_count); | |
| 2023 } | |
| 2024 | |
| 2025 __ jmp(L_copy_32_bytes); | |
| 2026 | |
| 2027 // Copy trailing qwords | |
| 2028 __ BIND(L_copy_8_bytes); | |
| 2029 __ movq(rax, Address(from, qword_count, Address::times_8, -8)); | |
| 2030 __ movq(Address(to, qword_count, Address::times_8, -8), rax); | |
| 2031 __ decrementq(qword_count); | |
| 2032 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 2033 | |
| 2034 if (is_oop) { | |
| 2035 __ jmp(L_exit); | |
| 2036 } else { | |
| 2037 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); | |
| 2038 restore_arg_regs(); | |
| 2039 __ xorq(rax, rax); // return 0 | |
| 2040 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 2041 __ ret(0); | |
| 2042 } | |
| 2043 | |
| 2044 // Copy in 32-bytes chunks | |
| 2045 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 2046 | |
| 2047 if (is_oop) { | |
| 2048 __ BIND(L_exit); | |
| 2049 __ leaq(rcx, Address(to, saved_count, Address::times_8, -8)); | |
| 2050 gen_write_ref_array_post_barrier(to, rcx, rax); | |
| 2051 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); | |
| 2052 } else { | |
| 2053 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); | |
| 2054 } | |
| 2055 restore_arg_regs(); | |
| 2056 __ xorq(rax, rax); // return 0 | |
| 2057 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 2058 __ ret(0); | |
| 2059 | |
| 2060 return start; | |
| 2061 } | |
| 2062 | |
| 2063 | |
| 2064 // Helper for generating a dynamic type check. | |
| 2065 // Smashes no registers. | |
| 2066 void generate_type_check(Register sub_klass, | |
| 2067 Register super_check_offset, | |
| 2068 Register super_klass, | |
| 2069 Label& L_success) { | |
| 2070 assert_different_registers(sub_klass, super_check_offset, super_klass); | |
| 2071 | |
| 2072 BLOCK_COMMENT("type_check:"); | |
| 2073 | |
| 2074 Label L_miss; | |
| 2075 | |
| 2076 // a couple of useful fields in sub_klass: | |
| 2077 int ss_offset = (klassOopDesc::header_size() * HeapWordSize + | |
| 2078 Klass::secondary_supers_offset_in_bytes()); | |
| 2079 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + | |
| 2080 Klass::secondary_super_cache_offset_in_bytes()); | |
| 2081 Address secondary_supers_addr(sub_klass, ss_offset); | |
| 2082 Address super_cache_addr( sub_klass, sc_offset); | |
| 2083 | |
| 2084 // if the pointers are equal, we are done (e.g., String[] elements) | |
| 2085 __ cmpq(super_klass, sub_klass); | |
| 2086 __ jcc(Assembler::equal, L_success); | |
| 2087 | |
| 2088 // check the supertype display: | |
| 2089 Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0); | |
| 2090 __ cmpq(super_klass, super_check_addr); // test the super type | |
| 2091 __ jcc(Assembler::equal, L_success); | |
| 2092 | |
| 2093 // if it was a primary super, we can just fail immediately | |
| 2094 __ cmpl(super_check_offset, sc_offset); | |
| 2095 __ jcc(Assembler::notEqual, L_miss); | |
| 2096 | |
| 2097 // Now do a linear scan of the secondary super-klass chain. | |
| 2098 // The repne_scan instruction uses fixed registers, which we must spill. | |
| 2099 // (We need a couple more temps in any case.) | |
| 2100 // This code is rarely used, so simplicity is a virtue here. | |
| 2101 inc_counter_np(SharedRuntime::_partial_subtype_ctr); | |
| 2102 { | |
| 2103 __ pushq(rax); | |
| 2104 __ pushq(rcx); | |
| 2105 __ pushq(rdi); | |
| 2106 assert_different_registers(sub_klass, super_klass, rax, rcx, rdi); | |
| 2107 | |
| 2108 __ movq(rdi, secondary_supers_addr); | |
| 2109 // Load the array length. | |
| 2110 __ movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes())); | |
| 2111 // Skip to start of data. | |
| 2112 __ addq(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); | |
| 2113 // Scan rcx words at [rdi] for occurance of rax | |
| 2114 // Set NZ/Z based on last compare | |
| 2115 __ movq(rax, super_klass); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2116 if (UseCompressedOops) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2117 // Compare against compressed form. Don't need to uncompress because |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2118 // looks like orig rax is restored in popq below. |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2119 __ encode_heap_oop(rax); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2120 __ repne_scanl(); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2121 } else { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2122 __ repne_scanq(); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2123 } |
| 0 | 2124 |
| 2125 // Unspill the temp. registers: | |
| 2126 __ popq(rdi); | |
| 2127 __ popq(rcx); | |
| 2128 __ popq(rax); | |
| 2129 | |
| 2130 __ jcc(Assembler::notEqual, L_miss); | |
| 2131 } | |
| 2132 | |
| 2133 // Success. Cache the super we found and proceed in triumph. | |
| 2134 __ movq(super_cache_addr, super_klass); // note: rax is dead | |
| 2135 __ jmp(L_success); | |
| 2136 | |
| 2137 // Fall through on failure! | |
| 2138 __ BIND(L_miss); | |
| 2139 } | |
| 2140 | |
| 2141 // | |
| 2142 // Generate checkcasting array copy stub | |
| 2143 // | |
| 2144 // Input: | |
| 2145 // c_rarg0 - source array address | |
| 2146 // c_rarg1 - destination array address | |
| 2147 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 2148 // c_rarg3 - size_t ckoff (super_check_offset) | |
| 2149 // not Win64 | |
| 2150 // c_rarg4 - oop ckval (super_klass) | |
| 2151 // Win64 | |
| 2152 // rsp+40 - oop ckval (super_klass) | |
| 2153 // | |
| 2154 // Output: | |
| 2155 // rax == 0 - success | |
| 2156 // rax == -1^K - failure, where K is partial transfer count | |
| 2157 // | |
| 2158 address generate_checkcast_copy(const char *name) { | |
| 2159 | |
| 2160 Label L_load_element, L_store_element, L_do_card_marks, L_done; | |
| 2161 | |
| 2162 // Input registers (after setup_arg_regs) | |
| 2163 const Register from = rdi; // source array address | |
| 2164 const Register to = rsi; // destination array address | |
| 2165 const Register length = rdx; // elements count | |
| 2166 const Register ckoff = rcx; // super_check_offset | |
| 2167 const Register ckval = r8; // super_klass | |
| 2168 | |
| 2169 // Registers used as temps (r13, r14 are save-on-entry) | |
| 2170 const Register end_from = from; // source array end address | |
| 2171 const Register end_to = r13; // destination array end address | |
| 2172 const Register count = rdx; // -(count_remaining) | |
| 2173 const Register r14_length = r14; // saved copy of length | |
| 2174 // End pointers are inclusive, and if length is not zero they point | |
| 2175 // to the last unit copied: end_to[0] := end_from[0] | |
| 2176 | |
| 2177 const Register rax_oop = rax; // actual oop copied | |
| 2178 const Register r11_klass = r11; // oop._klass | |
| 2179 | |
| 2180 //--------------------------------------------------------------- | |
| 2181 // Assembler stub will be used for this call to arraycopy | |
| 2182 // if the two arrays are subtypes of Object[] but the | |
| 2183 // destination array type is not equal to or a supertype | |
| 2184 // of the source type. Each element must be separately | |
| 2185 // checked. | |
| 2186 | |
| 2187 __ align(CodeEntryAlignment); | |
| 2188 StubCodeMark mark(this, "StubRoutines", name); | |
| 2189 address start = __ pc(); | |
| 2190 | |
| 2191 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 2192 | |
| 2193 checkcast_copy_entry = __ pc(); | |
| 2194 BLOCK_COMMENT("Entry:"); | |
| 2195 | |
| 2196 #ifdef ASSERT | |
| 2197 // caller guarantees that the arrays really are different | |
| 2198 // otherwise, we would have to make conjoint checks | |
| 2199 { Label L; | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2200 array_overlap_test(L, TIMES_OOP); |
| 0 | 2201 __ stop("checkcast_copy within a single array"); |
| 2202 __ bind(L); | |
| 2203 } | |
| 2204 #endif //ASSERT | |
| 2205 | |
| 2206 // allocate spill slots for r13, r14 | |
| 2207 enum { | |
| 2208 saved_r13_offset, | |
| 2209 saved_r14_offset, | |
| 2210 saved_rbp_offset, | |
| 2211 saved_rip_offset, | |
| 2212 saved_rarg0_offset | |
| 2213 }; | |
| 2214 __ subq(rsp, saved_rbp_offset * wordSize); | |
| 2215 __ movq(Address(rsp, saved_r13_offset * wordSize), r13); | |
| 2216 __ movq(Address(rsp, saved_r14_offset * wordSize), r14); | |
| 2217 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx | |
| 2218 // ckoff => rcx, ckval => r8 | |
| 2219 // r9 and r10 may be used to save non-volatile registers | |
| 2220 #ifdef _WIN64 | |
| 2221 // last argument (#4) is on stack on Win64 | |
| 2222 const int ckval_offset = saved_rarg0_offset + 4; | |
| 2223 __ movq(ckval, Address(rsp, ckval_offset * wordSize)); | |
| 2224 #endif | |
| 2225 | |
| 2226 // check that int operands are properly extended to size_t | |
| 2227 assert_clean_int(length, rax); | |
| 2228 assert_clean_int(ckoff, rax); | |
| 2229 | |
| 2230 #ifdef ASSERT | |
| 2231 BLOCK_COMMENT("assert consistent ckoff/ckval"); | |
| 2232 // The ckoff and ckval must be mutually consistent, | |
| 2233 // even though caller generates both. | |
| 2234 { Label L; | |
| 2235 int sco_offset = (klassOopDesc::header_size() * HeapWordSize + | |
| 2236 Klass::super_check_offset_offset_in_bytes()); | |
| 2237 __ cmpl(ckoff, Address(ckval, sco_offset)); | |
| 2238 __ jcc(Assembler::equal, L); | |
| 2239 __ stop("super_check_offset inconsistent"); | |
| 2240 __ bind(L); | |
| 2241 } | |
| 2242 #endif //ASSERT | |
| 2243 | |
| 2244 // Loop-invariant addresses. They are exclusive end pointers. | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2245 Address end_from_addr(from, length, TIMES_OOP, 0); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2246 Address end_to_addr(to, length, TIMES_OOP, 0); |
| 0 | 2247 // Loop-variant addresses. They assume post-incremented count < 0. |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2248 Address from_element_addr(end_from, count, TIMES_OOP, 0); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2249 Address to_element_addr(end_to, count, TIMES_OOP, 0); |
| 0 | 2250 |
| 2251 gen_write_ref_array_pre_barrier(to, count); | |
| 2252 | |
| 2253 // Copy from low to high addresses, indexed from the end of each array. | |
| 2254 __ leaq(end_from, end_from_addr); | |
| 2255 __ leaq(end_to, end_to_addr); | |
| 2256 __ movq(r14_length, length); // save a copy of the length | |
| 2257 assert(length == count, ""); // else fix next line: | |
| 2258 __ negq(count); // negate and test the length | |
| 2259 __ jcc(Assembler::notZero, L_load_element); | |
| 2260 | |
| 2261 // Empty array: Nothing to do. | |
| 2262 __ xorq(rax, rax); // return 0 on (trivial) success | |
| 2263 __ jmp(L_done); | |
| 2264 | |
| 2265 // ======== begin loop ======== | |
| 2266 // (Loop is rotated; its entry is L_load_element.) | |
| 2267 // Loop control: | |
| 2268 // for (count = -count; count != 0; count++) | |
| 2269 // Base pointers src, dst are biased by 8*(count-1),to last element. | |
| 2270 __ align(16); | |
| 2271 | |
| 2272 __ BIND(L_store_element); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2273 __ store_heap_oop(to_element_addr, rax_oop); // store the oop |
| 0 | 2274 __ incrementq(count); // increment the count toward zero |
| 2275 __ jcc(Assembler::zero, L_do_card_marks); | |
| 2276 | |
| 2277 // ======== loop entry is here ======== | |
| 2278 __ BIND(L_load_element); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2279 __ load_heap_oop(rax_oop, from_element_addr); // load the oop |
| 0 | 2280 __ testq(rax_oop, rax_oop); |
| 2281 __ jcc(Assembler::zero, L_store_element); | |
| 2282 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2283 __ load_klass(r11_klass, rax_oop);// query the object klass |
| 0 | 2284 generate_type_check(r11_klass, ckoff, ckval, L_store_element); |
| 2285 // ======== end loop ======== | |
| 2286 | |
| 2287 // It was a real error; we must depend on the caller to finish the job. | |
| 2288 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops. | |
| 2289 // Emit GC store barriers for the oops we have copied (r14 + rdx), | |
| 2290 // and report their number to the caller. | |
| 2291 assert_different_registers(rax, r14_length, count, to, end_to, rcx); | |
| 2292 __ leaq(end_to, to_element_addr); | |
| 2293 gen_write_ref_array_post_barrier(to, end_to, rcx); | |
| 2294 __ movq(rax, r14_length); // original oops | |
| 2295 __ addq(rax, count); // K = (original - remaining) oops | |
| 2296 __ notq(rax); // report (-1^K) to caller | |
| 2297 __ jmp(L_done); | |
| 2298 | |
| 2299 // Come here on success only. | |
| 2300 __ BIND(L_do_card_marks); | |
| 2301 __ addq(end_to, -wordSize); // make an inclusive end pointer | |
| 2302 gen_write_ref_array_post_barrier(to, end_to, rcx); | |
| 2303 __ xorq(rax, rax); // return 0 on success | |
| 2304 | |
| 2305 // Common exit point (success or failure). | |
| 2306 __ BIND(L_done); | |
| 2307 __ movq(r13, Address(rsp, saved_r13_offset * wordSize)); | |
| 2308 __ movq(r14, Address(rsp, saved_r14_offset * wordSize)); | |
| 2309 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); | |
| 2310 restore_arg_regs(); | |
| 2311 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 2312 __ ret(0); | |
| 2313 | |
| 2314 return start; | |
| 2315 } | |
| 2316 | |
| 2317 // | |
| 2318 // Generate 'unsafe' array copy stub | |
| 2319 // Though just as safe as the other stubs, it takes an unscaled | |
| 2320 // size_t argument instead of an element count. | |
| 2321 // | |
| 2322 // Input: | |
| 2323 // c_rarg0 - source array address | |
| 2324 // c_rarg1 - destination array address | |
| 2325 // c_rarg2 - byte count, treated as ssize_t, can be zero | |
| 2326 // | |
| 2327 // Examines the alignment of the operands and dispatches | |
| 2328 // to a long, int, short, or byte copy loop. | |
| 2329 // | |
| 2330 address generate_unsafe_copy(const char *name) { | |
| 2331 | |
| 2332 Label L_long_aligned, L_int_aligned, L_short_aligned; | |
| 2333 | |
| 2334 // Input registers (before setup_arg_regs) | |
| 2335 const Register from = c_rarg0; // source array address | |
| 2336 const Register to = c_rarg1; // destination array address | |
| 2337 const Register size = c_rarg2; // byte count (size_t) | |
| 2338 | |
| 2339 // Register used as a temp | |
| 2340 const Register bits = rax; // test copy of low bits | |
| 2341 | |
| 2342 __ align(CodeEntryAlignment); | |
| 2343 StubCodeMark mark(this, "StubRoutines", name); | |
| 2344 address start = __ pc(); | |
| 2345 | |
| 2346 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 2347 | |
| 2348 // bump this on entry, not on exit: | |
| 2349 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr); | |
| 2350 | |
| 2351 __ movq(bits, from); | |
| 2352 __ orq(bits, to); | |
| 2353 __ orq(bits, size); | |
| 2354 | |
| 2355 __ testb(bits, BytesPerLong-1); | |
| 2356 __ jccb(Assembler::zero, L_long_aligned); | |
| 2357 | |
| 2358 __ testb(bits, BytesPerInt-1); | |
| 2359 __ jccb(Assembler::zero, L_int_aligned); | |
| 2360 | |
| 2361 __ testb(bits, BytesPerShort-1); | |
| 2362 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry)); | |
| 2363 | |
| 2364 __ BIND(L_short_aligned); | |
| 2365 __ shrq(size, LogBytesPerShort); // size => short_count | |
| 2366 __ jump(RuntimeAddress(short_copy_entry)); | |
| 2367 | |
| 2368 __ BIND(L_int_aligned); | |
| 2369 __ shrq(size, LogBytesPerInt); // size => int_count | |
| 2370 __ jump(RuntimeAddress(int_copy_entry)); | |
| 2371 | |
| 2372 __ BIND(L_long_aligned); | |
| 2373 __ shrq(size, LogBytesPerLong); // size => qword_count | |
| 2374 __ jump(RuntimeAddress(long_copy_entry)); | |
| 2375 | |
| 2376 return start; | |
| 2377 } | |
| 2378 | |
| 2379 // Perform range checks on the proposed arraycopy. | |
| 2380 // Kills temp, but nothing else. | |
| 2381 // Also, clean the sign bits of src_pos and dst_pos. | |
| 2382 void arraycopy_range_checks(Register src, // source array oop (c_rarg0) | |
| 2383 Register src_pos, // source position (c_rarg1) | |
| 2384 Register dst, // destination array oo (c_rarg2) | |
| 2385 Register dst_pos, // destination position (c_rarg3) | |
| 2386 Register length, | |
| 2387 Register temp, | |
| 2388 Label& L_failed) { | |
| 2389 BLOCK_COMMENT("arraycopy_range_checks:"); | |
| 2390 | |
| 2391 // if (src_pos + length > arrayOop(src)->length()) FAIL; | |
| 2392 __ movl(temp, length); | |
| 2393 __ addl(temp, src_pos); // src_pos + length | |
| 2394 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes())); | |
| 2395 __ jcc(Assembler::above, L_failed); | |
| 2396 | |
| 2397 // if (dst_pos + length > arrayOop(dst)->length()) FAIL; | |
| 2398 __ movl(temp, length); | |
| 2399 __ addl(temp, dst_pos); // dst_pos + length | |
| 2400 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes())); | |
| 2401 __ jcc(Assembler::above, L_failed); | |
| 2402 | |
| 2403 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'. | |
| 2404 // Move with sign extension can be used since they are positive. | |
| 2405 __ movslq(src_pos, src_pos); | |
| 2406 __ movslq(dst_pos, dst_pos); | |
| 2407 | |
| 2408 BLOCK_COMMENT("arraycopy_range_checks done"); | |
| 2409 } | |
| 2410 | |
| 2411 // | |
| 2412 // Generate generic array copy stubs | |
| 2413 // | |
| 2414 // Input: | |
| 2415 // c_rarg0 - src oop | |
| 2416 // c_rarg1 - src_pos (32-bits) | |
| 2417 // c_rarg2 - dst oop | |
| 2418 // c_rarg3 - dst_pos (32-bits) | |
| 2419 // not Win64 | |
| 2420 // c_rarg4 - element count (32-bits) | |
| 2421 // Win64 | |
| 2422 // rsp+40 - element count (32-bits) | |
| 2423 // | |
| 2424 // Output: | |
| 2425 // rax == 0 - success | |
| 2426 // rax == -1^K - failure, where K is partial transfer count | |
| 2427 // | |
| 2428 address generate_generic_copy(const char *name) { | |
| 2429 | |
| 2430 Label L_failed, L_failed_0, L_objArray; | |
| 2431 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs; | |
| 2432 | |
| 2433 // Input registers | |
| 2434 const Register src = c_rarg0; // source array oop | |
| 2435 const Register src_pos = c_rarg1; // source position | |
| 2436 const Register dst = c_rarg2; // destination array oop | |
| 2437 const Register dst_pos = c_rarg3; // destination position | |
| 2438 // elements count is on stack on Win64 | |
| 2439 #ifdef _WIN64 | |
| 2440 #define C_RARG4 Address(rsp, 6 * wordSize) | |
| 2441 #else | |
| 2442 #define C_RARG4 c_rarg4 | |
| 2443 #endif | |
| 2444 | |
| 2445 { int modulus = CodeEntryAlignment; | |
| 2446 int target = modulus - 5; // 5 = sizeof jmp(L_failed) | |
| 2447 int advance = target - (__ offset() % modulus); | |
| 2448 if (advance < 0) advance += modulus; | |
| 2449 if (advance > 0) __ nop(advance); | |
| 2450 } | |
| 2451 StubCodeMark mark(this, "StubRoutines", name); | |
| 2452 | |
| 2453 // Short-hop target to L_failed. Makes for denser prologue code. | |
| 2454 __ BIND(L_failed_0); | |
| 2455 __ jmp(L_failed); | |
| 2456 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed"); | |
| 2457 | |
| 2458 __ align(CodeEntryAlignment); | |
| 2459 address start = __ pc(); | |
| 2460 | |
| 2461 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 2462 | |
| 2463 // bump this on entry, not on exit: | |
| 2464 inc_counter_np(SharedRuntime::_generic_array_copy_ctr); | |
| 2465 | |
| 2466 //----------------------------------------------------------------------- | |
| 2467 // Assembler stub will be used for this call to arraycopy | |
| 2468 // if the following conditions are met: | |
| 2469 // | |
| 2470 // (1) src and dst must not be null. | |
| 2471 // (2) src_pos must not be negative. | |
| 2472 // (3) dst_pos must not be negative. | |
| 2473 // (4) length must not be negative. | |
| 2474 // (5) src klass and dst klass should be the same and not NULL. | |
| 2475 // (6) src and dst should be arrays. | |
| 2476 // (7) src_pos + length must not exceed length of src. | |
| 2477 // (8) dst_pos + length must not exceed length of dst. | |
| 2478 // | |
| 2479 | |
| 2480 // if (src == NULL) return -1; | |
| 2481 __ testq(src, src); // src oop | |
| 2482 size_t j1off = __ offset(); | |
| 2483 __ jccb(Assembler::zero, L_failed_0); | |
| 2484 | |
| 2485 // if (src_pos < 0) return -1; | |
| 2486 __ testl(src_pos, src_pos); // src_pos (32-bits) | |
| 2487 __ jccb(Assembler::negative, L_failed_0); | |
| 2488 | |
| 2489 // if (dst == NULL) return -1; | |
| 2490 __ testq(dst, dst); // dst oop | |
| 2491 __ jccb(Assembler::zero, L_failed_0); | |
| 2492 | |
| 2493 // if (dst_pos < 0) return -1; | |
| 2494 __ testl(dst_pos, dst_pos); // dst_pos (32-bits) | |
| 2495 size_t j4off = __ offset(); | |
| 2496 __ jccb(Assembler::negative, L_failed_0); | |
| 2497 | |
| 2498 // The first four tests are very dense code, | |
| 2499 // but not quite dense enough to put four | |
| 2500 // jumps in a 16-byte instruction fetch buffer. | |
| 2501 // That's good, because some branch predicters | |
| 2502 // do not like jumps so close together. | |
| 2503 // Make sure of this. | |
| 2504 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps"); | |
| 2505 | |
| 2506 // registers used as temp | |
| 2507 const Register r11_length = r11; // elements count to copy | |
| 2508 const Register r10_src_klass = r10; // array klass | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2509 const Register r9_dst_klass = r9; // dest array klass |
| 0 | 2510 |
| 2511 // if (length < 0) return -1; | |
| 2512 __ movl(r11_length, C_RARG4); // length (elements count, 32-bits value) | |
| 2513 __ testl(r11_length, r11_length); | |
| 2514 __ jccb(Assembler::negative, L_failed_0); | |
| 2515 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2516 __ load_klass(r10_src_klass, src); |
| 0 | 2517 #ifdef ASSERT |
| 2518 // assert(src->klass() != NULL); | |
| 2519 BLOCK_COMMENT("assert klasses not null"); | |
| 2520 { Label L1, L2; | |
| 2521 __ testq(r10_src_klass, r10_src_klass); | |
| 2522 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL | |
| 2523 __ bind(L1); | |
| 2524 __ stop("broken null klass"); | |
| 2525 __ bind(L2); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2526 __ load_klass(r9_dst_klass, dst); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2527 __ cmpq(r9_dst_klass, 0); |
| 0 | 2528 __ jcc(Assembler::equal, L1); // this would be broken also |
| 2529 BLOCK_COMMENT("assert done"); | |
| 2530 } | |
| 2531 #endif | |
| 2532 | |
| 2533 // Load layout helper (32-bits) | |
| 2534 // | |
| 2535 // |array_tag| | header_size | element_type | |log2_element_size| | |
| 2536 // 32 30 24 16 8 2 0 | |
| 2537 // | |
| 2538 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 | |
| 2539 // | |
| 2540 | |
| 2541 int lh_offset = klassOopDesc::header_size() * HeapWordSize + | |
| 2542 Klass::layout_helper_offset_in_bytes(); | |
| 2543 | |
| 2544 const Register rax_lh = rax; // layout helper | |
| 2545 | |
| 2546 __ movl(rax_lh, Address(r10_src_klass, lh_offset)); | |
| 2547 | |
| 2548 // Handle objArrays completely differently... | |
| 2549 jint objArray_lh = Klass::array_layout_helper(T_OBJECT); | |
| 2550 __ cmpl(rax_lh, objArray_lh); | |
| 2551 __ jcc(Assembler::equal, L_objArray); | |
| 2552 | |
| 2553 // if (src->klass() != dst->klass()) return -1; | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2554 __ load_klass(r9_dst_klass, dst); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2555 __ cmpq(r10_src_klass, r9_dst_klass); |
| 0 | 2556 __ jcc(Assembler::notEqual, L_failed); |
| 2557 | |
| 2558 // if (!src->is_Array()) return -1; | |
| 2559 __ cmpl(rax_lh, Klass::_lh_neutral_value); | |
| 2560 __ jcc(Assembler::greaterEqual, L_failed); | |
| 2561 | |
| 2562 // At this point, it is known to be a typeArray (array_tag 0x3). | |
| 2563 #ifdef ASSERT | |
| 2564 { Label L; | |
| 2565 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift)); | |
| 2566 __ jcc(Assembler::greaterEqual, L); | |
| 2567 __ stop("must be a primitive array"); | |
| 2568 __ bind(L); | |
| 2569 } | |
| 2570 #endif | |
| 2571 | |
| 2572 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, | |
| 2573 r10, L_failed); | |
| 2574 | |
| 2575 // typeArrayKlass | |
| 2576 // | |
| 2577 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); | |
| 2578 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); | |
| 2579 // | |
| 2580 | |
| 2581 const Register r10_offset = r10; // array offset | |
| 2582 const Register rax_elsize = rax_lh; // element size | |
| 2583 | |
| 2584 __ movl(r10_offset, rax_lh); | |
| 2585 __ shrl(r10_offset, Klass::_lh_header_size_shift); | |
| 2586 __ andq(r10_offset, Klass::_lh_header_size_mask); // array_offset | |
| 2587 __ addq(src, r10_offset); // src array offset | |
| 2588 __ addq(dst, r10_offset); // dst array offset | |
| 2589 BLOCK_COMMENT("choose copy loop based on element size"); | |
| 2590 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize | |
| 2591 | |
| 2592 // next registers should be set before the jump to corresponding stub | |
| 2593 const Register from = c_rarg0; // source array address | |
| 2594 const Register to = c_rarg1; // destination array address | |
| 2595 const Register count = c_rarg2; // elements count | |
| 2596 | |
| 2597 // 'from', 'to', 'count' registers should be set in such order | |
| 2598 // since they are the same as 'src', 'src_pos', 'dst'. | |
| 2599 | |
| 2600 __ BIND(L_copy_bytes); | |
| 2601 __ cmpl(rax_elsize, 0); | |
| 2602 __ jccb(Assembler::notEqual, L_copy_shorts); | |
| 2603 __ leaq(from, Address(src, src_pos, Address::times_1, 0));// src_addr | |
| 2604 __ leaq(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr | |
| 2605 __ movslq(count, r11_length); // length | |
| 2606 __ jump(RuntimeAddress(byte_copy_entry)); | |
| 2607 | |
| 2608 __ BIND(L_copy_shorts); | |
| 2609 __ cmpl(rax_elsize, LogBytesPerShort); | |
| 2610 __ jccb(Assembler::notEqual, L_copy_ints); | |
| 2611 __ leaq(from, Address(src, src_pos, Address::times_2, 0));// src_addr | |
| 2612 __ leaq(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr | |
| 2613 __ movslq(count, r11_length); // length | |
| 2614 __ jump(RuntimeAddress(short_copy_entry)); | |
| 2615 | |
| 2616 __ BIND(L_copy_ints); | |
| 2617 __ cmpl(rax_elsize, LogBytesPerInt); | |
| 2618 __ jccb(Assembler::notEqual, L_copy_longs); | |
| 2619 __ leaq(from, Address(src, src_pos, Address::times_4, 0));// src_addr | |
| 2620 __ leaq(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr | |
| 2621 __ movslq(count, r11_length); // length | |
| 2622 __ jump(RuntimeAddress(int_copy_entry)); | |
| 2623 | |
| 2624 __ BIND(L_copy_longs); | |
| 2625 #ifdef ASSERT | |
| 2626 { Label L; | |
| 2627 __ cmpl(rax_elsize, LogBytesPerLong); | |
| 2628 __ jcc(Assembler::equal, L); | |
| 2629 __ stop("must be long copy, but elsize is wrong"); | |
| 2630 __ bind(L); | |
| 2631 } | |
| 2632 #endif | |
| 2633 __ leaq(from, Address(src, src_pos, Address::times_8, 0));// src_addr | |
| 2634 __ leaq(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr | |
| 2635 __ movslq(count, r11_length); // length | |
| 2636 __ jump(RuntimeAddress(long_copy_entry)); | |
| 2637 | |
| 2638 // objArrayKlass | |
| 2639 __ BIND(L_objArray); | |
| 2640 // live at this point: r10_src_klass, src[_pos], dst[_pos] | |
| 2641 | |
| 2642 Label L_plain_copy, L_checkcast_copy; | |
| 2643 // test array classes for subtyping | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2644 __ load_klass(r9_dst_klass, dst); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2645 __ cmpq(r10_src_klass, r9_dst_klass); // usual case is exact equality |
| 0 | 2646 __ jcc(Assembler::notEqual, L_checkcast_copy); |
| 2647 | |
| 2648 // Identically typed arrays can be copied without element-wise checks. | |
| 2649 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, | |
| 2650 r10, L_failed); | |
| 2651 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2652 __ leaq(from, Address(src, src_pos, TIMES_OOP, |
| 0 | 2653 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2654 __ leaq(to, Address(dst, dst_pos, TIMES_OOP, |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2655 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr |
| 0 | 2656 __ movslq(count, r11_length); // length |
| 2657 __ BIND(L_plain_copy); | |
| 2658 __ jump(RuntimeAddress(oop_copy_entry)); | |
| 2659 | |
| 2660 __ BIND(L_checkcast_copy); | |
| 2661 // live at this point: r10_src_klass, !r11_length | |
| 2662 { | |
| 2663 // assert(r11_length == C_RARG4); // will reload from here | |
| 2664 Register r11_dst_klass = r11; | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2665 __ load_klass(r11_dst_klass, dst); |
| 0 | 2666 |
| 2667 // Before looking at dst.length, make sure dst is also an objArray. | |
| 2668 __ cmpl(Address(r11_dst_klass, lh_offset), objArray_lh); | |
| 2669 __ jcc(Assembler::notEqual, L_failed); | |
| 2670 | |
| 2671 // It is safe to examine both src.length and dst.length. | |
| 2672 #ifndef _WIN64 | |
| 2673 arraycopy_range_checks(src, src_pos, dst, dst_pos, C_RARG4, | |
| 2674 rax, L_failed); | |
| 2675 #else | |
| 2676 __ movl(r11_length, C_RARG4); // reload | |
| 2677 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, | |
| 2678 rax, L_failed); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2679 __ load_klass(r11_dst_klass, dst); // reload |
| 0 | 2680 #endif |
| 2681 | |
| 2682 // Marshal the base address arguments now, freeing registers. | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2683 __ leaq(from, Address(src, src_pos, TIMES_OOP, |
| 0 | 2684 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2685 __ leaq(to, Address(dst, dst_pos, TIMES_OOP, |
| 0 | 2686 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); |
| 2687 __ movl(count, C_RARG4); // length (reloaded) | |
| 2688 Register sco_temp = c_rarg3; // this register is free now | |
| 2689 assert_different_registers(from, to, count, sco_temp, | |
| 2690 r11_dst_klass, r10_src_klass); | |
| 2691 assert_clean_int(count, sco_temp); | |
| 2692 | |
| 2693 // Generate the type check. | |
| 2694 int sco_offset = (klassOopDesc::header_size() * HeapWordSize + | |
| 2695 Klass::super_check_offset_offset_in_bytes()); | |
| 2696 __ movl(sco_temp, Address(r11_dst_klass, sco_offset)); | |
| 2697 assert_clean_int(sco_temp, rax); | |
| 2698 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy); | |
| 2699 | |
| 2700 // Fetch destination element klass from the objArrayKlass header. | |
| 2701 int ek_offset = (klassOopDesc::header_size() * HeapWordSize + | |
| 2702 objArrayKlass::element_klass_offset_in_bytes()); | |
| 2703 __ movq(r11_dst_klass, Address(r11_dst_klass, ek_offset)); | |
| 2704 __ movl(sco_temp, Address(r11_dst_klass, sco_offset)); | |
| 2705 assert_clean_int(sco_temp, rax); | |
| 2706 | |
| 2707 // the checkcast_copy loop needs two extra arguments: | |
| 2708 assert(c_rarg3 == sco_temp, "#3 already in place"); | |
| 2709 __ movq(C_RARG4, r11_dst_klass); // dst.klass.element_klass | |
| 2710 __ jump(RuntimeAddress(checkcast_copy_entry)); | |
| 2711 } | |
| 2712 | |
| 2713 __ BIND(L_failed); | |
| 2714 __ xorq(rax, rax); | |
| 2715 __ notq(rax); // return -1 | |
| 2716 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 2717 __ ret(0); | |
| 2718 | |
| 2719 return start; | |
| 2720 } | |
| 2721 | |
| 2722 #undef length_arg | |
| 2723 | |
| 2724 void generate_arraycopy_stubs() { | |
| 2725 // Call the conjoint generation methods immediately after | |
| 2726 // the disjoint ones so that short branches from the former | |
| 2727 // to the latter can be generated. | |
| 2728 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy"); | |
| 2729 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, "jbyte_arraycopy"); | |
| 2730 | |
| 2731 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy"); | |
| 2732 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, "jshort_arraycopy"); | |
| 2733 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2734 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, "jint_disjoint_arraycopy"); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2735 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, "jint_arraycopy"); |
| 0 | 2736 |
| 2737 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, "jlong_disjoint_arraycopy"); | |
| 2738 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, "jlong_arraycopy"); | |
| 2739 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2740 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2741 if (UseCompressedOops) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2742 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, "oop_disjoint_arraycopy"); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2743 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, "oop_arraycopy"); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2744 } else { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2745 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, "oop_disjoint_arraycopy"); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2746 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, "oop_arraycopy"); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2747 } |
| 0 | 2748 |
| 2749 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy"); | |
| 2750 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy"); | |
| 2751 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy"); | |
| 2752 | |
| 2753 // We don't generate specialized code for HeapWord-aligned source | |
| 2754 // arrays, so just use the code we've already generated | |
| 2755 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy; | |
| 2756 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy; | |
| 2757 | |
| 2758 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy; | |
| 2759 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy; | |
| 2760 | |
| 2761 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy; | |
| 2762 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy; | |
| 2763 | |
| 2764 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy; | |
| 2765 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy; | |
| 2766 | |
| 2767 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy; | |
| 2768 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; | |
| 2769 } | |
| 2770 | |
| 2771 #undef __ | |
| 2772 #define __ masm-> | |
| 2773 | |
| 2774 // Continuation point for throwing of implicit exceptions that are | |
| 2775 // not handled in the current activation. Fabricates an exception | |
| 2776 // oop and initiates normal exception dispatching in this | |
| 2777 // frame. Since we need to preserve callee-saved values (currently | |
| 2778 // only for C2, but done for C1 as well) we need a callee-saved oop | |
| 2779 // map and therefore have to make these stubs into RuntimeStubs | |
| 2780 // rather than BufferBlobs. If the compiler needs all registers to | |
| 2781 // be preserved between the fault point and the exception handler | |
| 2782 // then it must assume responsibility for that in | |
| 2783 // AbstractCompiler::continuation_for_implicit_null_exception or | |
| 2784 // continuation_for_implicit_division_by_zero_exception. All other | |
| 2785 // implicit exceptions (e.g., NullPointerException or | |
| 2786 // AbstractMethodError on entry) are either at call sites or | |
| 2787 // otherwise assume that stack unwinding will be initiated, so | |
| 2788 // caller saved registers were assumed volatile in the compiler. | |
| 2789 address generate_throw_exception(const char* name, | |
| 2790 address runtime_entry, | |
| 2791 bool restore_saved_exception_pc) { | |
| 2792 // Information about frame layout at time of blocking runtime call. | |
| 2793 // Note that we only have to preserve callee-saved registers since | |
| 2794 // the compilers are responsible for supplying a continuation point | |
| 2795 // if they expect all registers to be preserved. | |
| 2796 enum layout { | |
| 2797 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt, | |
| 2798 rbp_off2, | |
| 2799 return_off, | |
| 2800 return_off2, | |
| 2801 framesize // inclusive of return address | |
| 2802 }; | |
| 2803 | |
| 2804 int insts_size = 512; | |
| 2805 int locs_size = 64; | |
| 2806 | |
| 2807 CodeBuffer code(name, insts_size, locs_size); | |
| 2808 OopMapSet* oop_maps = new OopMapSet(); | |
| 2809 MacroAssembler* masm = new MacroAssembler(&code); | |
| 2810 | |
| 2811 address start = __ pc(); | |
| 2812 | |
| 2813 // This is an inlined and slightly modified version of call_VM | |
| 2814 // which has the ability to fetch the return PC out of | |
| 2815 // thread-local storage and also sets up last_Java_sp slightly | |
| 2816 // differently than the real call_VM | |
| 2817 if (restore_saved_exception_pc) { | |
| 2818 __ movq(rax, | |
| 2819 Address(r15_thread, | |
| 2820 in_bytes(JavaThread::saved_exception_pc_offset()))); | |
| 2821 __ pushq(rax); | |
| 2822 } | |
| 2823 | |
| 2824 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 2825 | |
| 2826 assert(is_even(framesize/2), "sp not 16-byte aligned"); | |
| 2827 | |
| 2828 // return address and rbp are already in place | |
| 2829 __ subq(rsp, (framesize-4) << LogBytesPerInt); // prolog | |
| 2830 | |
| 2831 int frame_complete = __ pc() - start; | |
| 2832 | |
| 2833 // Set up last_Java_sp and last_Java_fp | |
| 2834 __ set_last_Java_frame(rsp, rbp, NULL); | |
| 2835 | |
| 2836 // Call runtime | |
| 2837 __ movq(c_rarg0, r15_thread); | |
| 2838 BLOCK_COMMENT("call runtime_entry"); | |
| 2839 __ call(RuntimeAddress(runtime_entry)); | |
| 2840 | |
| 2841 // Generate oop map | |
| 2842 OopMap* map = new OopMap(framesize, 0); | |
| 2843 | |
| 2844 oop_maps->add_gc_map(__ pc() - start, map); | |
| 2845 | |
| 2846 __ reset_last_Java_frame(true, false); | |
| 2847 | |
| 2848 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 2849 | |
| 2850 // check for pending exceptions | |
| 2851 #ifdef ASSERT | |
| 2852 Label L; | |
| 2853 __ cmpq(Address(r15_thread, Thread::pending_exception_offset()), | |
| 2854 (int) NULL); | |
| 2855 __ jcc(Assembler::notEqual, L); | |
| 2856 __ should_not_reach_here(); | |
| 2857 __ bind(L); | |
| 2858 #endif // ASSERT | |
| 2859 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); | |
| 2860 | |
| 2861 | |
| 2862 // codeBlob framesize is in words (not VMRegImpl::slot_size) | |
| 2863 RuntimeStub* stub = | |
| 2864 RuntimeStub::new_runtime_stub(name, | |
| 2865 &code, | |
| 2866 frame_complete, | |
| 2867 (framesize >> (LogBytesPerWord - LogBytesPerInt)), | |
| 2868 oop_maps, false); | |
| 2869 return stub->entry_point(); | |
| 2870 } | |
| 2871 | |
| 2872 // Initialization | |
| 2873 void generate_initial() { | |
| 2874 // Generates all stubs and initializes the entry points | |
| 2875 | |
| 2876 // This platform-specific stub is needed by generate_call_stub() | |
| 2877 StubRoutines::amd64::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80); | |
| 2878 | |
| 2879 // entry points that exist in all platforms Note: This is code | |
| 2880 // that could be shared among different platforms - however the | |
| 2881 // benefit seems to be smaller than the disadvantage of having a | |
| 2882 // much more complicated generator structure. See also comment in | |
| 2883 // stubRoutines.hpp. | |
| 2884 | |
| 2885 StubRoutines::_forward_exception_entry = generate_forward_exception(); | |
| 2886 | |
| 2887 StubRoutines::_call_stub_entry = | |
| 2888 generate_call_stub(StubRoutines::_call_stub_return_address); | |
| 2889 | |
| 2890 // is referenced by megamorphic call | |
| 2891 StubRoutines::_catch_exception_entry = generate_catch_exception(); | |
| 2892 | |
| 2893 // atomic calls | |
| 2894 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); | |
| 2895 StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr(); | |
| 2896 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg(); | |
| 2897 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long(); | |
| 2898 StubRoutines::_atomic_add_entry = generate_atomic_add(); | |
| 2899 StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr(); | |
| 2900 StubRoutines::_fence_entry = generate_orderaccess_fence(); | |
| 2901 | |
| 2902 StubRoutines::_handler_for_unsafe_access_entry = | |
| 2903 generate_handler_for_unsafe_access(); | |
| 2904 | |
| 2905 // platform dependent | |
| 2906 StubRoutines::amd64::_get_previous_fp_entry = generate_get_previous_fp(); | |
| 2907 | |
| 2908 StubRoutines::amd64::_verify_mxcsr_entry = generate_verify_mxcsr(); | |
| 2909 } | |
| 2910 | |
| 2911 void generate_all() { | |
| 2912 // Generates all stubs and initializes the entry points | |
| 2913 | |
| 2914 // These entry points require SharedInfo::stack0 to be set up in | |
| 2915 // non-core builds and need to be relocatable, so they each | |
| 2916 // fabricate a RuntimeStub internally. | |
| 2917 StubRoutines::_throw_AbstractMethodError_entry = | |
| 2918 generate_throw_exception("AbstractMethodError throw_exception", | |
| 2919 CAST_FROM_FN_PTR(address, | |
| 2920 SharedRuntime:: | |
| 2921 throw_AbstractMethodError), | |
| 2922 false); | |
| 2923 | |
|
16
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2924 StubRoutines::_throw_IncompatibleClassChangeError_entry = |
|
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2925 generate_throw_exception("IncompatibleClassChangeError throw_exception", |
|
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2926 CAST_FROM_FN_PTR(address, |
|
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2927 SharedRuntime:: |
|
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2928 throw_IncompatibleClassChangeError), |
|
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2929 false); |
|
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2930 |
| 0 | 2931 StubRoutines::_throw_ArithmeticException_entry = |
| 2932 generate_throw_exception("ArithmeticException throw_exception", | |
| 2933 CAST_FROM_FN_PTR(address, | |
| 2934 SharedRuntime:: | |
| 2935 throw_ArithmeticException), | |
| 2936 true); | |
| 2937 | |
| 2938 StubRoutines::_throw_NullPointerException_entry = | |
| 2939 generate_throw_exception("NullPointerException throw_exception", | |
| 2940 CAST_FROM_FN_PTR(address, | |
| 2941 SharedRuntime:: | |
| 2942 throw_NullPointerException), | |
| 2943 true); | |
| 2944 | |
| 2945 StubRoutines::_throw_NullPointerException_at_call_entry = | |
| 2946 generate_throw_exception("NullPointerException at call throw_exception", | |
| 2947 CAST_FROM_FN_PTR(address, | |
| 2948 SharedRuntime:: | |
| 2949 throw_NullPointerException_at_call), | |
| 2950 false); | |
| 2951 | |
| 2952 StubRoutines::_throw_StackOverflowError_entry = | |
| 2953 generate_throw_exception("StackOverflowError throw_exception", | |
| 2954 CAST_FROM_FN_PTR(address, | |
| 2955 SharedRuntime:: | |
| 2956 throw_StackOverflowError), | |
| 2957 false); | |
| 2958 | |
| 2959 // entry points that are platform specific | |
| 2960 StubRoutines::amd64::_f2i_fixup = generate_f2i_fixup(); | |
| 2961 StubRoutines::amd64::_f2l_fixup = generate_f2l_fixup(); | |
| 2962 StubRoutines::amd64::_d2i_fixup = generate_d2i_fixup(); | |
| 2963 StubRoutines::amd64::_d2l_fixup = generate_d2l_fixup(); | |
| 2964 | |
| 2965 StubRoutines::amd64::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF); | |
| 2966 StubRoutines::amd64::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000); | |
| 2967 StubRoutines::amd64::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF); | |
| 2968 StubRoutines::amd64::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000); | |
| 2969 | |
| 2970 // support for verify_oop (must happen after universe_init) | |
| 2971 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop(); | |
| 2972 | |
| 2973 // arraycopy stubs used by compilers | |
| 2974 generate_arraycopy_stubs(); | |
| 2975 } | |
| 2976 | |
| 2977 public: | |
| 2978 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { | |
| 2979 if (all) { | |
| 2980 generate_all(); | |
| 2981 } else { | |
| 2982 generate_initial(); | |
| 2983 } | |
| 2984 } | |
| 2985 }; // end class declaration | |
| 2986 | |
| 2987 address StubGenerator::disjoint_byte_copy_entry = NULL; | |
| 2988 address StubGenerator::disjoint_short_copy_entry = NULL; | |
| 2989 address StubGenerator::disjoint_int_copy_entry = NULL; | |
| 2990 address StubGenerator::disjoint_long_copy_entry = NULL; | |
| 2991 address StubGenerator::disjoint_oop_copy_entry = NULL; | |
| 2992 | |
| 2993 address StubGenerator::byte_copy_entry = NULL; | |
| 2994 address StubGenerator::short_copy_entry = NULL; | |
| 2995 address StubGenerator::int_copy_entry = NULL; | |
| 2996 address StubGenerator::long_copy_entry = NULL; | |
| 2997 address StubGenerator::oop_copy_entry = NULL; | |
| 2998 | |
| 2999 address StubGenerator::checkcast_copy_entry = NULL; | |
| 3000 | |
| 3001 void StubGenerator_generate(CodeBuffer* code, bool all) { | |
| 3002 StubGenerator g(code, all); | |
| 3003 } |