Mercurial > hg > truffle
annotate src/cpu/x86/vm/stubGenerator_x86_64.cpp @ 113:ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
Summary: Compressed oops in instances, arrays, and headers. Code contributors are coleenp, phh, never, swamyv
Reviewed-by: jmasa, kamg, acorn, tbell, kvn, rasbold
| author | coleenp |
|---|---|
| date | Sun, 13 Apr 2008 17:43:42 -0400 |
| parents | f8236e79048a |
| children | b130b98db9cf |
| rev | line source |
|---|---|
| 0 | 1 /* |
| 2 * Copyright 2003-2007 Sun Microsystems, Inc. All Rights Reserved. | |
| 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
| 4 * | |
| 5 * This code is free software; you can redistribute it and/or modify it | |
| 6 * under the terms of the GNU General Public License version 2 only, as | |
| 7 * published by the Free Software Foundation. | |
| 8 * | |
| 9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
| 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
| 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
| 12 * version 2 for more details (a copy is included in the LICENSE file that | |
| 13 * accompanied this code). | |
| 14 * | |
| 15 * You should have received a copy of the GNU General Public License version | |
| 16 * 2 along with this work; if not, write to the Free Software Foundation, | |
| 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
| 18 * | |
| 19 * Please contact 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 | |
| 916 // [tos + 2]: saved flags | |
| 917 // [tos + 3]: return address | |
| 918 // * [tos + 4]: error message (char*) | |
| 919 // * [tos + 5]: object to verify (oop) | |
| 920 // * [tos + 6]: saved rax - saved by caller and bashed | |
| 921 // * = popped on exit | |
| 922 address generate_verify_oop() { | |
| 923 StubCodeMark mark(this, "StubRoutines", "verify_oop"); | |
| 924 address start = __ pc(); | |
| 925 | |
| 926 Label exit, error; | |
| 927 | |
| 928 __ pushfq(); | |
| 929 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr())); | |
| 930 | |
| 931 // save c_rarg2 and c_rarg3 | |
| 932 __ pushq(c_rarg2); | |
| 933 __ pushq(c_rarg3); | |
| 934 | |
| 935 // get object | |
| 936 __ movq(rax, Address(rsp, 5 * wordSize)); | |
| 937 | |
| 938 // make sure object is 'reasonable' | |
| 939 __ testq(rax, rax); | |
| 940 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK | |
| 941 // Check if the oop is in the right area of memory | |
| 942 __ movq(c_rarg2, rax); | |
| 943 __ movptr(c_rarg3, (int64_t) Universe::verify_oop_mask()); | |
| 944 __ andq(c_rarg2, c_rarg3); | |
| 945 __ movptr(c_rarg3, (int64_t) Universe::verify_oop_bits()); | |
| 946 __ cmpq(c_rarg2, c_rarg3); | |
| 947 __ jcc(Assembler::notZero, error); | |
| 948 | |
| 949 // 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
|
950 __ load_klass(rax, rax); // get klass |
| 0 | 951 __ testq(rax, rax); |
| 952 __ jcc(Assembler::zero, error); // if klass is NULL it is broken | |
| 953 // Check if the klass is in the right area of memory | |
| 954 __ movq(c_rarg2, rax); | |
| 955 __ movptr(c_rarg3, (int64_t) Universe::verify_klass_mask()); | |
| 956 __ andq(c_rarg2, c_rarg3); | |
| 957 __ movptr(c_rarg3, (int64_t) Universe::verify_klass_bits()); | |
| 958 __ cmpq(c_rarg2, c_rarg3); | |
| 959 __ jcc(Assembler::notZero, error); | |
| 960 | |
| 961 // 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
|
962 __ load_klass(rax, rax); |
| 0 | 963 __ testq(rax, rax); |
| 964 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken | |
| 965 // Check if the klass' klass is in the right area of memory | |
| 966 __ movptr(c_rarg3, (int64_t) Universe::verify_klass_mask()); | |
| 967 __ andq(rax, c_rarg3); | |
| 968 __ movptr(c_rarg3, (int64_t) Universe::verify_klass_bits()); | |
| 969 __ cmpq(rax, c_rarg3); | |
| 970 __ jcc(Assembler::notZero, error); | |
| 971 | |
| 972 // return if everything seems ok | |
| 973 __ bind(exit); | |
| 974 __ movq(rax, Address(rsp, 6 * wordSize)); // get saved rax back | |
| 975 __ popq(c_rarg3); // restore c_rarg3 | |
| 976 __ popq(c_rarg2); // restore c_rarg2 | |
| 977 __ popfq(); // restore flags | |
| 978 __ ret(3 * wordSize); // pop caller saved stuff | |
| 979 | |
| 980 // handle errors | |
| 981 __ bind(error); | |
| 982 __ movq(rax, Address(rsp, 6 * wordSize)); // get saved rax back | |
| 983 __ popq(c_rarg3); // get saved c_rarg3 back | |
| 984 __ popq(c_rarg2); // get saved c_rarg2 back | |
| 985 __ popfq(); // get saved flags off stack -- | |
| 986 // will be ignored | |
| 987 | |
| 988 __ pushaq(); // push registers | |
| 989 // (rip is already | |
| 990 // already pushed) | |
| 991 // debug(char* msg, int64_t regs[]) | |
| 992 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and | |
| 993 // pushed all the registers, so now the stack looks like: | |
| 994 // [tos + 0] 16 saved registers | |
| 995 // [tos + 16] return address | |
| 996 // [tos + 17] error message (char*) | |
| 997 | |
| 998 __ movq(c_rarg0, Address(rsp, 17 * wordSize)); // pass address of error message | |
| 999 __ movq(c_rarg1, rsp); // pass address of regs on stack | |
| 1000 __ movq(r12, rsp); // remember rsp | |
| 1001 __ subq(rsp, frame::arg_reg_save_area_bytes);// windows | |
| 1002 __ andq(rsp, -16); // align stack as required by ABI | |
| 1003 BLOCK_COMMENT("call MacroAssembler::debug"); | |
| 1004 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug))); | |
| 1005 __ movq(rsp, r12); // restore rsp | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1006 __ reinit_heapbase(); // r12 is heapbase |
| 0 | 1007 __ popaq(); // pop registers |
| 1008 __ ret(3 * wordSize); // pop caller saved stuff | |
| 1009 | |
| 1010 return start; | |
| 1011 } | |
| 1012 | |
| 1013 static address disjoint_byte_copy_entry; | |
| 1014 static address disjoint_short_copy_entry; | |
| 1015 static address disjoint_int_copy_entry; | |
| 1016 static address disjoint_long_copy_entry; | |
| 1017 static address disjoint_oop_copy_entry; | |
| 1018 | |
| 1019 static address byte_copy_entry; | |
| 1020 static address short_copy_entry; | |
| 1021 static address int_copy_entry; | |
| 1022 static address long_copy_entry; | |
| 1023 static address oop_copy_entry; | |
| 1024 | |
| 1025 static address checkcast_copy_entry; | |
| 1026 | |
| 1027 // | |
| 1028 // Verify that a register contains clean 32-bits positive value | |
| 1029 // (high 32-bits are 0) so it could be used in 64-bits shifts. | |
| 1030 // | |
| 1031 // Input: | |
| 1032 // Rint - 32-bits value | |
| 1033 // Rtmp - scratch | |
| 1034 // | |
| 1035 void assert_clean_int(Register Rint, Register Rtmp) { | |
| 1036 #ifdef ASSERT | |
| 1037 Label L; | |
| 1038 assert_different_registers(Rtmp, Rint); | |
| 1039 __ movslq(Rtmp, Rint); | |
| 1040 __ cmpq(Rtmp, Rint); | |
| 1041 __ jccb(Assembler::equal, L); | |
| 1042 __ stop("high 32-bits of int value are not 0"); | |
| 1043 __ bind(L); | |
| 1044 #endif | |
| 1045 } | |
| 1046 | |
| 1047 // Generate overlap test for array copy stubs | |
| 1048 // | |
| 1049 // Input: | |
| 1050 // c_rarg0 - from | |
| 1051 // c_rarg1 - to | |
| 1052 // c_rarg2 - element count | |
| 1053 // | |
| 1054 // Output: | |
| 1055 // rax - &from[element count - 1] | |
| 1056 // | |
| 1057 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) { | |
| 1058 assert(no_overlap_target != NULL, "must be generated"); | |
| 1059 array_overlap_test(no_overlap_target, NULL, sf); | |
| 1060 } | |
| 1061 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) { | |
| 1062 array_overlap_test(NULL, &L_no_overlap, sf); | |
| 1063 } | |
| 1064 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) { | |
| 1065 const Register from = c_rarg0; | |
| 1066 const Register to = c_rarg1; | |
| 1067 const Register count = c_rarg2; | |
| 1068 const Register end_from = rax; | |
| 1069 | |
| 1070 __ cmpq(to, from); | |
| 1071 __ leaq(end_from, Address(from, count, sf, 0)); | |
| 1072 if (NOLp == NULL) { | |
| 1073 ExternalAddress no_overlap(no_overlap_target); | |
| 1074 __ jump_cc(Assembler::belowEqual, no_overlap); | |
| 1075 __ cmpq(to, end_from); | |
| 1076 __ jump_cc(Assembler::aboveEqual, no_overlap); | |
| 1077 } else { | |
| 1078 __ jcc(Assembler::belowEqual, (*NOLp)); | |
| 1079 __ cmpq(to, end_from); | |
| 1080 __ jcc(Assembler::aboveEqual, (*NOLp)); | |
| 1081 } | |
| 1082 } | |
| 1083 | |
| 1084 // Shuffle first three arg regs on Windows into Linux/Solaris locations. | |
| 1085 // | |
| 1086 // Outputs: | |
| 1087 // rdi - rcx | |
| 1088 // rsi - rdx | |
| 1089 // rdx - r8 | |
| 1090 // rcx - r9 | |
| 1091 // | |
| 1092 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter | |
| 1093 // are non-volatile. r9 and r10 should not be used by the caller. | |
| 1094 // | |
| 1095 void setup_arg_regs(int nargs = 3) { | |
| 1096 const Register saved_rdi = r9; | |
| 1097 const Register saved_rsi = r10; | |
| 1098 assert(nargs == 3 || nargs == 4, "else fix"); | |
| 1099 #ifdef _WIN64 | |
| 1100 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9, | |
| 1101 "unexpected argument registers"); | |
| 1102 if (nargs >= 4) | |
| 1103 __ movq(rax, r9); // r9 is also saved_rdi | |
| 1104 __ movq(saved_rdi, rdi); | |
| 1105 __ movq(saved_rsi, rsi); | |
| 1106 __ movq(rdi, rcx); // c_rarg0 | |
| 1107 __ movq(rsi, rdx); // c_rarg1 | |
| 1108 __ movq(rdx, r8); // c_rarg2 | |
| 1109 if (nargs >= 4) | |
| 1110 __ movq(rcx, rax); // c_rarg3 (via rax) | |
| 1111 #else | |
| 1112 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx, | |
| 1113 "unexpected argument registers"); | |
| 1114 #endif | |
| 1115 } | |
| 1116 | |
| 1117 void restore_arg_regs() { | |
| 1118 const Register saved_rdi = r9; | |
| 1119 const Register saved_rsi = r10; | |
| 1120 #ifdef _WIN64 | |
| 1121 __ movq(rdi, saved_rdi); | |
| 1122 __ movq(rsi, saved_rsi); | |
| 1123 #endif | |
| 1124 } | |
| 1125 | |
| 1126 // Generate code for an array write pre barrier | |
| 1127 // | |
| 1128 // addr - starting address | |
| 1129 // count - element count | |
| 1130 // | |
| 1131 // Destroy no registers! | |
| 1132 // | |
| 1133 void gen_write_ref_array_pre_barrier(Register addr, Register count) { | |
| 1134 #if 0 // G1 - only | |
| 1135 assert_different_registers(addr, c_rarg1); | |
| 1136 assert_different_registers(count, c_rarg0); | |
| 1137 BarrierSet* bs = Universe::heap()->barrier_set(); | |
| 1138 switch (bs->kind()) { | |
| 1139 case BarrierSet::G1SATBCT: | |
| 1140 case BarrierSet::G1SATBCTLogging: | |
| 1141 { | |
| 1142 __ pushaq(); // push registers | |
| 1143 __ movq(c_rarg0, addr); | |
| 1144 __ movq(c_rarg1, count); | |
| 1145 __ call(RuntimeAddress(BarrierSet::static_write_ref_array_pre)); | |
| 1146 __ popaq(); | |
| 1147 } | |
| 1148 break; | |
| 1149 case BarrierSet::CardTableModRef: | |
| 1150 case BarrierSet::CardTableExtension: | |
| 1151 case BarrierSet::ModRef: | |
| 1152 break; | |
| 1153 default : | |
| 1154 ShouldNotReachHere(); | |
| 1155 | |
| 1156 } | |
| 1157 #endif // 0 G1 - only | |
| 1158 } | |
| 1159 | |
| 1160 // | |
| 1161 // Generate code for an array write post barrier | |
| 1162 // | |
| 1163 // Input: | |
| 1164 // start - register containing starting address of destination array | |
| 1165 // end - register containing ending address of destination array | |
| 1166 // scratch - scratch register | |
| 1167 // | |
| 1168 // The input registers are overwritten. | |
| 1169 // The ending address is inclusive. | |
| 1170 void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) { | |
| 1171 assert_different_registers(start, end, scratch); | |
| 1172 BarrierSet* bs = Universe::heap()->barrier_set(); | |
| 1173 switch (bs->kind()) { | |
| 1174 #if 0 // G1 - only | |
| 1175 case BarrierSet::G1SATBCT: | |
| 1176 case BarrierSet::G1SATBCTLogging: | |
| 1177 | |
| 1178 { | |
| 1179 __ pushaq(); // push registers (overkill) | |
| 1180 // must compute element count unless barrier set interface is changed (other platforms supply count) | |
| 1181 assert_different_registers(start, end, scratch); | |
| 1182 __ leaq(scratch, Address(end, wordSize)); | |
| 1183 __ subq(scratch, start); | |
| 1184 __ shrq(scratch, LogBytesPerWord); | |
| 1185 __ movq(c_rarg0, start); | |
| 1186 __ movq(c_rarg1, scratch); | |
| 1187 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post)); | |
| 1188 __ popaq(); | |
| 1189 } | |
| 1190 break; | |
| 1191 #endif // 0 G1 - only | |
| 1192 case BarrierSet::CardTableModRef: | |
| 1193 case BarrierSet::CardTableExtension: | |
| 1194 { | |
| 1195 CardTableModRefBS* ct = (CardTableModRefBS*)bs; | |
| 1196 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); | |
| 1197 | |
| 1198 Label L_loop; | |
| 1199 | |
| 1200 __ shrq(start, CardTableModRefBS::card_shift); | |
| 1201 __ shrq(end, CardTableModRefBS::card_shift); | |
| 1202 __ subq(end, start); // number of bytes to copy | |
| 1203 | |
| 1204 const Register count = end; // 'end' register contains bytes count now | |
| 1205 __ lea(scratch, ExternalAddress((address)ct->byte_map_base)); | |
| 1206 __ addq(start, scratch); | |
| 1207 __ BIND(L_loop); | |
| 1208 __ movb(Address(start, count, Address::times_1), 0); | |
| 1209 __ decrementq(count); | |
| 1210 __ jcc(Assembler::greaterEqual, L_loop); | |
| 1211 } | |
| 1212 } | |
| 1213 } | |
| 1214 | |
| 1215 // Copy big chunks forward | |
| 1216 // | |
| 1217 // Inputs: | |
| 1218 // end_from - source arrays end address | |
| 1219 // end_to - destination array end address | |
| 1220 // qword_count - 64-bits element count, negative | |
| 1221 // to - scratch | |
| 1222 // L_copy_32_bytes - entry label | |
| 1223 // L_copy_8_bytes - exit label | |
| 1224 // | |
| 1225 void copy_32_bytes_forward(Register end_from, Register end_to, | |
| 1226 Register qword_count, Register to, | |
| 1227 Label& L_copy_32_bytes, Label& L_copy_8_bytes) { | |
| 1228 DEBUG_ONLY(__ stop("enter at entry label, not here")); | |
| 1229 Label L_loop; | |
| 1230 __ align(16); | |
| 1231 __ BIND(L_loop); | |
| 1232 __ movq(to, Address(end_from, qword_count, Address::times_8, -24)); | |
| 1233 __ movq(Address(end_to, qword_count, Address::times_8, -24), to); | |
| 1234 __ movq(to, Address(end_from, qword_count, Address::times_8, -16)); | |
| 1235 __ movq(Address(end_to, qword_count, Address::times_8, -16), to); | |
| 1236 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8)); | |
| 1237 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to); | |
| 1238 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0)); | |
| 1239 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to); | |
| 1240 __ BIND(L_copy_32_bytes); | |
| 1241 __ addq(qword_count, 4); | |
| 1242 __ jcc(Assembler::lessEqual, L_loop); | |
| 1243 __ subq(qword_count, 4); | |
| 1244 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords | |
| 1245 } | |
| 1246 | |
| 1247 | |
| 1248 // Copy big chunks backward | |
| 1249 // | |
| 1250 // Inputs: | |
| 1251 // from - source arrays address | |
| 1252 // dest - destination array address | |
| 1253 // qword_count - 64-bits element count | |
| 1254 // to - scratch | |
| 1255 // L_copy_32_bytes - entry label | |
| 1256 // L_copy_8_bytes - exit label | |
| 1257 // | |
| 1258 void copy_32_bytes_backward(Register from, Register dest, | |
| 1259 Register qword_count, Register to, | |
| 1260 Label& L_copy_32_bytes, Label& L_copy_8_bytes) { | |
| 1261 DEBUG_ONLY(__ stop("enter at entry label, not here")); | |
| 1262 Label L_loop; | |
| 1263 __ align(16); | |
| 1264 __ BIND(L_loop); | |
| 1265 __ movq(to, Address(from, qword_count, Address::times_8, 24)); | |
| 1266 __ movq(Address(dest, qword_count, Address::times_8, 24), to); | |
| 1267 __ movq(to, Address(from, qword_count, Address::times_8, 16)); | |
| 1268 __ movq(Address(dest, qword_count, Address::times_8, 16), to); | |
| 1269 __ movq(to, Address(from, qword_count, Address::times_8, 8)); | |
| 1270 __ movq(Address(dest, qword_count, Address::times_8, 8), to); | |
| 1271 __ movq(to, Address(from, qword_count, Address::times_8, 0)); | |
| 1272 __ movq(Address(dest, qword_count, Address::times_8, 0), to); | |
| 1273 __ BIND(L_copy_32_bytes); | |
| 1274 __ subq(qword_count, 4); | |
| 1275 __ jcc(Assembler::greaterEqual, L_loop); | |
| 1276 __ addq(qword_count, 4); | |
| 1277 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords | |
| 1278 } | |
| 1279 | |
| 1280 | |
| 1281 // Arguments: | |
| 1282 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary | |
| 1283 // ignored | |
| 1284 // name - stub name string | |
| 1285 // | |
| 1286 // Inputs: | |
| 1287 // c_rarg0 - source array address | |
| 1288 // c_rarg1 - destination array address | |
| 1289 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1290 // | |
| 1291 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries, | |
| 1292 // we let the hardware handle it. The one to eight bytes within words, | |
| 1293 // dwords or qwords that span cache line boundaries will still be loaded | |
| 1294 // and stored atomically. | |
| 1295 // | |
| 1296 // Side Effects: | |
| 1297 // disjoint_byte_copy_entry is set to the no-overlap entry point | |
| 1298 // used by generate_conjoint_byte_copy(). | |
| 1299 // | |
| 1300 address generate_disjoint_byte_copy(bool aligned, const char *name) { | |
| 1301 __ align(CodeEntryAlignment); | |
| 1302 StubCodeMark mark(this, "StubRoutines", name); | |
| 1303 address start = __ pc(); | |
| 1304 | |
| 1305 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes; | |
| 1306 Label L_copy_byte, L_exit; | |
| 1307 const Register from = rdi; // source array address | |
| 1308 const Register to = rsi; // destination array address | |
| 1309 const Register count = rdx; // elements count | |
| 1310 const Register byte_count = rcx; | |
| 1311 const Register qword_count = count; | |
| 1312 const Register end_from = from; // source array end address | |
| 1313 const Register end_to = to; // destination array end address | |
| 1314 // End pointers are inclusive, and if count is not zero they point | |
| 1315 // to the last unit copied: end_to[0] := end_from[0] | |
| 1316 | |
| 1317 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1318 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1319 | |
| 1320 disjoint_byte_copy_entry = __ pc(); | |
| 1321 BLOCK_COMMENT("Entry:"); | |
| 1322 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1323 | |
| 1324 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1325 // r9 and r10 may be used to save non-volatile registers | |
| 1326 | |
| 1327 // 'from', 'to' and 'count' are now valid | |
| 1328 __ movq(byte_count, count); | |
| 1329 __ shrq(count, 3); // count => qword_count | |
| 1330 | |
| 1331 // Copy from low to high addresses. Use 'to' as scratch. | |
| 1332 __ leaq(end_from, Address(from, qword_count, Address::times_8, -8)); | |
| 1333 __ leaq(end_to, Address(to, qword_count, Address::times_8, -8)); | |
| 1334 __ negq(qword_count); // make the count negative | |
| 1335 __ jmp(L_copy_32_bytes); | |
| 1336 | |
| 1337 // Copy trailing qwords | |
| 1338 __ BIND(L_copy_8_bytes); | |
| 1339 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); | |
| 1340 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); | |
| 1341 __ incrementq(qword_count); | |
| 1342 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1343 | |
| 1344 // Check for and copy trailing dword | |
| 1345 __ BIND(L_copy_4_bytes); | |
| 1346 __ testq(byte_count, 4); | |
| 1347 __ jccb(Assembler::zero, L_copy_2_bytes); | |
| 1348 __ movl(rax, Address(end_from, 8)); | |
| 1349 __ movl(Address(end_to, 8), rax); | |
| 1350 | |
| 1351 __ addq(end_from, 4); | |
| 1352 __ addq(end_to, 4); | |
| 1353 | |
| 1354 // Check for and copy trailing word | |
| 1355 __ BIND(L_copy_2_bytes); | |
| 1356 __ testq(byte_count, 2); | |
| 1357 __ jccb(Assembler::zero, L_copy_byte); | |
| 1358 __ movw(rax, Address(end_from, 8)); | |
| 1359 __ movw(Address(end_to, 8), rax); | |
| 1360 | |
| 1361 __ addq(end_from, 2); | |
| 1362 __ addq(end_to, 2); | |
| 1363 | |
| 1364 // Check for and copy trailing byte | |
| 1365 __ BIND(L_copy_byte); | |
| 1366 __ testq(byte_count, 1); | |
| 1367 __ jccb(Assembler::zero, L_exit); | |
| 1368 __ movb(rax, Address(end_from, 8)); | |
| 1369 __ movb(Address(end_to, 8), rax); | |
| 1370 | |
| 1371 __ BIND(L_exit); | |
| 1372 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); | |
| 1373 restore_arg_regs(); | |
| 1374 __ xorq(rax, rax); // return 0 | |
| 1375 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1376 __ ret(0); | |
| 1377 | |
| 1378 // Copy in 32-bytes chunks | |
| 1379 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1380 __ jmp(L_copy_4_bytes); | |
| 1381 | |
| 1382 return start; | |
| 1383 } | |
| 1384 | |
| 1385 // Arguments: | |
| 1386 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary | |
| 1387 // ignored | |
| 1388 // name - stub name string | |
| 1389 // | |
| 1390 // Inputs: | |
| 1391 // c_rarg0 - source array address | |
| 1392 // c_rarg1 - destination array address | |
| 1393 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1394 // | |
| 1395 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries, | |
| 1396 // we let the hardware handle it. The one to eight bytes within words, | |
| 1397 // dwords or qwords that span cache line boundaries will still be loaded | |
| 1398 // and stored atomically. | |
| 1399 // | |
| 1400 address generate_conjoint_byte_copy(bool aligned, const char *name) { | |
| 1401 __ align(CodeEntryAlignment); | |
| 1402 StubCodeMark mark(this, "StubRoutines", name); | |
| 1403 address start = __ pc(); | |
| 1404 | |
| 1405 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes; | |
| 1406 const Register from = rdi; // source array address | |
| 1407 const Register to = rsi; // destination array address | |
| 1408 const Register count = rdx; // elements count | |
| 1409 const Register byte_count = rcx; | |
| 1410 const Register qword_count = count; | |
| 1411 | |
| 1412 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1413 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1414 | |
| 1415 byte_copy_entry = __ pc(); | |
| 1416 BLOCK_COMMENT("Entry:"); | |
| 1417 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1418 | |
| 1419 array_overlap_test(disjoint_byte_copy_entry, Address::times_1); | |
| 1420 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1421 // r9 and r10 may be used to save non-volatile registers | |
| 1422 | |
| 1423 // 'from', 'to' and 'count' are now valid | |
| 1424 __ movq(byte_count, count); | |
| 1425 __ shrq(count, 3); // count => qword_count | |
| 1426 | |
| 1427 // Copy from high to low addresses. | |
| 1428 | |
| 1429 // Check for and copy trailing byte | |
| 1430 __ testq(byte_count, 1); | |
| 1431 __ jcc(Assembler::zero, L_copy_2_bytes); | |
| 1432 __ movb(rax, Address(from, byte_count, Address::times_1, -1)); | |
| 1433 __ movb(Address(to, byte_count, Address::times_1, -1), rax); | |
| 1434 __ decrementq(byte_count); // Adjust for possible trailing word | |
| 1435 | |
| 1436 // Check for and copy trailing word | |
| 1437 __ BIND(L_copy_2_bytes); | |
| 1438 __ testq(byte_count, 2); | |
| 1439 __ jcc(Assembler::zero, L_copy_4_bytes); | |
| 1440 __ movw(rax, Address(from, byte_count, Address::times_1, -2)); | |
| 1441 __ movw(Address(to, byte_count, Address::times_1, -2), rax); | |
| 1442 | |
| 1443 // Check for and copy trailing dword | |
| 1444 __ BIND(L_copy_4_bytes); | |
| 1445 __ testq(byte_count, 4); | |
| 1446 __ jcc(Assembler::zero, L_copy_32_bytes); | |
| 1447 __ movl(rax, Address(from, qword_count, Address::times_8)); | |
| 1448 __ movl(Address(to, qword_count, Address::times_8), rax); | |
| 1449 __ jmp(L_copy_32_bytes); | |
| 1450 | |
| 1451 // Copy trailing qwords | |
| 1452 __ BIND(L_copy_8_bytes); | |
| 1453 __ movq(rax, Address(from, qword_count, Address::times_8, -8)); | |
| 1454 __ movq(Address(to, qword_count, Address::times_8, -8), rax); | |
| 1455 __ decrementq(qword_count); | |
| 1456 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1457 | |
| 1458 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); | |
| 1459 restore_arg_regs(); | |
| 1460 __ xorq(rax, rax); // return 0 | |
| 1461 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1462 __ ret(0); | |
| 1463 | |
| 1464 // Copy in 32-bytes chunks | |
| 1465 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1466 | |
| 1467 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); | |
| 1468 restore_arg_regs(); | |
| 1469 __ xorq(rax, rax); // return 0 | |
| 1470 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1471 __ ret(0); | |
| 1472 | |
| 1473 return start; | |
| 1474 } | |
| 1475 | |
| 1476 // Arguments: | |
| 1477 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary | |
| 1478 // ignored | |
| 1479 // name - stub name string | |
| 1480 // | |
| 1481 // Inputs: | |
| 1482 // c_rarg0 - source array address | |
| 1483 // c_rarg1 - destination array address | |
| 1484 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1485 // | |
| 1486 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we | |
| 1487 // let the hardware handle it. The two or four words within dwords | |
| 1488 // or qwords that span cache line boundaries will still be loaded | |
| 1489 // and stored atomically. | |
| 1490 // | |
| 1491 // Side Effects: | |
| 1492 // disjoint_short_copy_entry is set to the no-overlap entry point | |
| 1493 // used by generate_conjoint_short_copy(). | |
| 1494 // | |
| 1495 address generate_disjoint_short_copy(bool aligned, const char *name) { | |
| 1496 __ align(CodeEntryAlignment); | |
| 1497 StubCodeMark mark(this, "StubRoutines", name); | |
| 1498 address start = __ pc(); | |
| 1499 | |
| 1500 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit; | |
| 1501 const Register from = rdi; // source array address | |
| 1502 const Register to = rsi; // destination array address | |
| 1503 const Register count = rdx; // elements count | |
| 1504 const Register word_count = rcx; | |
| 1505 const Register qword_count = count; | |
| 1506 const Register end_from = from; // source array end address | |
| 1507 const Register end_to = to; // destination array end address | |
| 1508 // End pointers are inclusive, and if count is not zero they point | |
| 1509 // to the last unit copied: end_to[0] := end_from[0] | |
| 1510 | |
| 1511 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1512 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1513 | |
| 1514 disjoint_short_copy_entry = __ pc(); | |
| 1515 BLOCK_COMMENT("Entry:"); | |
| 1516 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1517 | |
| 1518 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1519 // r9 and r10 may be used to save non-volatile registers | |
| 1520 | |
| 1521 // 'from', 'to' and 'count' are now valid | |
| 1522 __ movq(word_count, count); | |
| 1523 __ shrq(count, 2); // count => qword_count | |
| 1524 | |
| 1525 // Copy from low to high addresses. Use 'to' as scratch. | |
| 1526 __ leaq(end_from, Address(from, qword_count, Address::times_8, -8)); | |
| 1527 __ leaq(end_to, Address(to, qword_count, Address::times_8, -8)); | |
| 1528 __ negq(qword_count); | |
| 1529 __ jmp(L_copy_32_bytes); | |
| 1530 | |
| 1531 // Copy trailing qwords | |
| 1532 __ BIND(L_copy_8_bytes); | |
| 1533 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); | |
| 1534 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); | |
| 1535 __ incrementq(qword_count); | |
| 1536 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1537 | |
| 1538 // Original 'dest' is trashed, so we can't use it as a | |
| 1539 // base register for a possible trailing word copy | |
| 1540 | |
| 1541 // Check for and copy trailing dword | |
| 1542 __ BIND(L_copy_4_bytes); | |
| 1543 __ testq(word_count, 2); | |
| 1544 __ jccb(Assembler::zero, L_copy_2_bytes); | |
| 1545 __ movl(rax, Address(end_from, 8)); | |
| 1546 __ movl(Address(end_to, 8), rax); | |
| 1547 | |
| 1548 __ addq(end_from, 4); | |
| 1549 __ addq(end_to, 4); | |
| 1550 | |
| 1551 // Check for and copy trailing word | |
| 1552 __ BIND(L_copy_2_bytes); | |
| 1553 __ testq(word_count, 1); | |
| 1554 __ jccb(Assembler::zero, L_exit); | |
| 1555 __ movw(rax, Address(end_from, 8)); | |
| 1556 __ movw(Address(end_to, 8), rax); | |
| 1557 | |
| 1558 __ BIND(L_exit); | |
| 1559 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); | |
| 1560 restore_arg_regs(); | |
| 1561 __ xorq(rax, rax); // return 0 | |
| 1562 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1563 __ ret(0); | |
| 1564 | |
| 1565 // Copy in 32-bytes chunks | |
| 1566 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1567 __ jmp(L_copy_4_bytes); | |
| 1568 | |
| 1569 return start; | |
| 1570 } | |
| 1571 | |
| 1572 // Arguments: | |
| 1573 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary | |
| 1574 // ignored | |
| 1575 // name - stub name string | |
| 1576 // | |
| 1577 // Inputs: | |
| 1578 // c_rarg0 - source array address | |
| 1579 // c_rarg1 - destination array address | |
| 1580 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1581 // | |
| 1582 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we | |
| 1583 // let the hardware handle it. The two or four words within dwords | |
| 1584 // or qwords that span cache line boundaries will still be loaded | |
| 1585 // and stored atomically. | |
| 1586 // | |
| 1587 address generate_conjoint_short_copy(bool aligned, const char *name) { | |
| 1588 __ align(CodeEntryAlignment); | |
| 1589 StubCodeMark mark(this, "StubRoutines", name); | |
| 1590 address start = __ pc(); | |
| 1591 | |
| 1592 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes; | |
| 1593 const Register from = rdi; // source array address | |
| 1594 const Register to = rsi; // destination array address | |
| 1595 const Register count = rdx; // elements count | |
| 1596 const Register word_count = rcx; | |
| 1597 const Register qword_count = count; | |
| 1598 | |
| 1599 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1600 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1601 | |
| 1602 short_copy_entry = __ pc(); | |
| 1603 BLOCK_COMMENT("Entry:"); | |
| 1604 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1605 | |
| 1606 array_overlap_test(disjoint_short_copy_entry, Address::times_2); | |
| 1607 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1608 // r9 and r10 may be used to save non-volatile registers | |
| 1609 | |
| 1610 // 'from', 'to' and 'count' are now valid | |
| 1611 __ movq(word_count, count); | |
| 1612 __ shrq(count, 2); // count => qword_count | |
| 1613 | |
| 1614 // Copy from high to low addresses. Use 'to' as scratch. | |
| 1615 | |
| 1616 // Check for and copy trailing word | |
| 1617 __ testq(word_count, 1); | |
| 1618 __ jccb(Assembler::zero, L_copy_4_bytes); | |
| 1619 __ movw(rax, Address(from, word_count, Address::times_2, -2)); | |
| 1620 __ movw(Address(to, word_count, Address::times_2, -2), rax); | |
| 1621 | |
| 1622 // Check for and copy trailing dword | |
| 1623 __ BIND(L_copy_4_bytes); | |
| 1624 __ testq(word_count, 2); | |
| 1625 __ jcc(Assembler::zero, L_copy_32_bytes); | |
| 1626 __ movl(rax, Address(from, qword_count, Address::times_8)); | |
| 1627 __ movl(Address(to, qword_count, Address::times_8), rax); | |
| 1628 __ jmp(L_copy_32_bytes); | |
| 1629 | |
| 1630 // Copy trailing qwords | |
| 1631 __ BIND(L_copy_8_bytes); | |
| 1632 __ movq(rax, Address(from, qword_count, Address::times_8, -8)); | |
| 1633 __ movq(Address(to, qword_count, Address::times_8, -8), rax); | |
| 1634 __ decrementq(qword_count); | |
| 1635 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1636 | |
| 1637 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); | |
| 1638 restore_arg_regs(); | |
| 1639 __ xorq(rax, rax); // return 0 | |
| 1640 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1641 __ ret(0); | |
| 1642 | |
| 1643 // Copy in 32-bytes chunks | |
| 1644 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1645 | |
| 1646 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); | |
| 1647 restore_arg_regs(); | |
| 1648 __ xorq(rax, rax); // return 0 | |
| 1649 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1650 __ ret(0); | |
| 1651 | |
| 1652 return start; | |
| 1653 } | |
| 1654 | |
| 1655 // Arguments: | |
| 1656 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary | |
| 1657 // ignored | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1658 // is_oop - true => oop array, so generate store check code |
| 0 | 1659 // name - stub name string |
| 1660 // | |
| 1661 // Inputs: | |
| 1662 // c_rarg0 - source array address | |
| 1663 // c_rarg1 - destination array address | |
| 1664 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1665 // | |
| 1666 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let | |
| 1667 // the hardware handle it. The two dwords within qwords that span | |
| 1668 // cache line boundaries will still be loaded and stored atomicly. | |
| 1669 // | |
| 1670 // Side Effects: | |
| 1671 // 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
|
1672 // used by generate_conjoint_int_oop_copy(). |
| 0 | 1673 // |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1674 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) { |
| 0 | 1675 __ align(CodeEntryAlignment); |
| 1676 StubCodeMark mark(this, "StubRoutines", name); | |
| 1677 address start = __ pc(); | |
| 1678 | |
| 1679 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit; | |
| 1680 const Register from = rdi; // source array address | |
| 1681 const Register to = rsi; // destination array address | |
| 1682 const Register count = rdx; // elements count | |
| 1683 const Register dword_count = rcx; | |
| 1684 const Register qword_count = count; | |
| 1685 const Register end_from = from; // source array end address | |
| 1686 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
|
1687 const Register saved_to = r11; // saved destination array address |
| 0 | 1688 // End pointers are inclusive, and if count is not zero they point |
| 1689 // to the last unit copied: end_to[0] := end_from[0] | |
| 1690 | |
| 1691 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1692 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1693 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1694 (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
|
1695 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1696 if (is_oop) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1697 // 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
|
1698 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
|
1699 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1700 |
| 0 | 1701 BLOCK_COMMENT("Entry:"); |
| 1702 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1703 | |
| 1704 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1705 // r9 and r10 may be used to save non-volatile registers | |
| 1706 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1707 if (is_oop) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1708 __ movq(saved_to, to); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1709 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1710 |
| 0 | 1711 // 'from', 'to' and 'count' are now valid |
| 1712 __ movq(dword_count, count); | |
| 1713 __ shrq(count, 1); // count => qword_count | |
| 1714 | |
| 1715 // Copy from low to high addresses. Use 'to' as scratch. | |
| 1716 __ leaq(end_from, Address(from, qword_count, Address::times_8, -8)); | |
| 1717 __ leaq(end_to, Address(to, qword_count, Address::times_8, -8)); | |
| 1718 __ negq(qword_count); | |
| 1719 __ jmp(L_copy_32_bytes); | |
| 1720 | |
| 1721 // Copy trailing qwords | |
| 1722 __ BIND(L_copy_8_bytes); | |
| 1723 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); | |
| 1724 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); | |
| 1725 __ incrementq(qword_count); | |
| 1726 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1727 | |
| 1728 // Check for and copy trailing dword | |
| 1729 __ BIND(L_copy_4_bytes); | |
| 1730 __ testq(dword_count, 1); // Only byte test since the value is 0 or 1 | |
| 1731 __ jccb(Assembler::zero, L_exit); | |
| 1732 __ movl(rax, Address(end_from, 8)); | |
| 1733 __ movl(Address(end_to, 8), rax); | |
| 1734 | |
| 1735 __ BIND(L_exit); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1736 if (is_oop) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1737 __ 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
|
1738 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
|
1739 } |
| 0 | 1740 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); |
| 1741 restore_arg_regs(); | |
| 1742 __ xorq(rax, rax); // return 0 | |
| 1743 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1744 __ ret(0); | |
| 1745 | |
| 1746 // Copy 32-bytes chunks | |
| 1747 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1748 __ jmp(L_copy_4_bytes); | |
| 1749 | |
| 1750 return start; | |
| 1751 } | |
| 1752 | |
| 1753 // Arguments: | |
| 1754 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary | |
| 1755 // ignored | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1756 // is_oop - true => oop array, so generate store check code |
| 0 | 1757 // name - stub name string |
| 1758 // | |
| 1759 // Inputs: | |
| 1760 // c_rarg0 - source array address | |
| 1761 // c_rarg1 - destination array address | |
| 1762 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1763 // | |
| 1764 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let | |
| 1765 // the hardware handle it. The two dwords within qwords that span | |
| 1766 // cache line boundaries will still be loaded and stored atomicly. | |
| 1767 // | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1768 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) { |
| 0 | 1769 __ align(CodeEntryAlignment); |
| 1770 StubCodeMark mark(this, "StubRoutines", name); | |
| 1771 address start = __ pc(); | |
| 1772 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1773 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit; |
| 0 | 1774 const Register from = rdi; // source array address |
| 1775 const Register to = rsi; // destination array address | |
| 1776 const Register count = rdx; // elements count | |
| 1777 const Register dword_count = rcx; | |
| 1778 const Register qword_count = count; | |
| 1779 | |
| 1780 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1781 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1782 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1783 if (is_oop) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1784 // 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
|
1785 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
|
1786 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1787 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1788 (is_oop ? oop_copy_entry : int_copy_entry) = __ pc(); |
| 0 | 1789 BLOCK_COMMENT("Entry:"); |
| 1790 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1791 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1792 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
|
1793 Address::times_4); |
| 0 | 1794 setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
| 1795 // r9 and r10 may be used to save non-volatile registers | |
| 1796 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1797 assert_clean_int(count, rax); // Make sure 'count' is clean int. |
| 0 | 1798 // 'from', 'to' and 'count' are now valid |
| 1799 __ movq(dword_count, count); | |
| 1800 __ shrq(count, 1); // count => qword_count | |
| 1801 | |
| 1802 // Copy from high to low addresses. Use 'to' as scratch. | |
| 1803 | |
| 1804 // Check for and copy trailing dword | |
| 1805 __ testq(dword_count, 1); | |
| 1806 __ jcc(Assembler::zero, L_copy_32_bytes); | |
| 1807 __ movl(rax, Address(from, dword_count, Address::times_4, -4)); | |
| 1808 __ movl(Address(to, dword_count, Address::times_4, -4), rax); | |
| 1809 __ jmp(L_copy_32_bytes); | |
| 1810 | |
| 1811 // Copy trailing qwords | |
| 1812 __ BIND(L_copy_8_bytes); | |
| 1813 __ movq(rax, Address(from, qword_count, Address::times_8, -8)); | |
| 1814 __ movq(Address(to, qword_count, Address::times_8, -8), rax); | |
| 1815 __ decrementq(qword_count); | |
| 1816 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1817 | |
| 1818 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
|
1819 if (is_oop) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1820 __ jmp(L_exit); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1821 } |
| 0 | 1822 restore_arg_regs(); |
| 1823 __ xorq(rax, rax); // return 0 | |
| 1824 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1825 __ ret(0); | |
| 1826 | |
| 1827 // Copy in 32-bytes chunks | |
| 1828 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1829 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1830 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
|
1831 __ bind(L_exit); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1832 if (is_oop) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1833 Register end_to = rdx; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1834 __ 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
|
1835 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
|
1836 } |
| 0 | 1837 restore_arg_regs(); |
| 1838 __ xorq(rax, rax); // return 0 | |
| 1839 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1840 __ ret(0); | |
| 1841 | |
| 1842 return start; | |
| 1843 } | |
| 1844 | |
| 1845 // Arguments: | |
| 1846 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes | |
| 1847 // ignored | |
| 1848 // is_oop - true => oop array, so generate store check code | |
| 1849 // name - stub name string | |
| 1850 // | |
| 1851 // Inputs: | |
| 1852 // c_rarg0 - source array address | |
| 1853 // c_rarg1 - destination array address | |
| 1854 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1855 // | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1856 // Side Effects: |
| 0 | 1857 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the |
| 1858 // no-overlap entry point used by generate_conjoint_long_oop_copy(). | |
| 1859 // | |
| 1860 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) { | |
| 1861 __ align(CodeEntryAlignment); | |
| 1862 StubCodeMark mark(this, "StubRoutines", name); | |
| 1863 address start = __ pc(); | |
| 1864 | |
| 1865 Label L_copy_32_bytes, L_copy_8_bytes, L_exit; | |
| 1866 const Register from = rdi; // source array address | |
| 1867 const Register to = rsi; // destination array address | |
| 1868 const Register qword_count = rdx; // elements count | |
| 1869 const Register end_from = from; // source array end address | |
| 1870 const Register end_to = rcx; // destination array end address | |
| 1871 const Register saved_to = to; | |
| 1872 // End pointers are inclusive, and if count is not zero they point | |
| 1873 // to the last unit copied: end_to[0] := end_from[0] | |
| 1874 | |
| 1875 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1876 // Save no-overlap entry point for generate_conjoint_long_oop_copy() | |
| 1877 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1878 | |
| 1879 if (is_oop) { | |
| 1880 disjoint_oop_copy_entry = __ pc(); | |
| 1881 // no registers are destroyed by this call | |
| 1882 gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2); | |
| 1883 } else { | |
| 1884 disjoint_long_copy_entry = __ pc(); | |
| 1885 } | |
| 1886 BLOCK_COMMENT("Entry:"); | |
| 1887 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1888 | |
| 1889 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1890 // r9 and r10 may be used to save non-volatile registers | |
| 1891 | |
| 1892 // 'from', 'to' and 'qword_count' are now valid | |
| 1893 | |
| 1894 // Copy from low to high addresses. Use 'to' as scratch. | |
| 1895 __ 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
|
1896 __ leaq(end_to, Address(to, qword_count, Address::times_8, -8)); |
| 0 | 1897 __ negq(qword_count); |
| 1898 __ jmp(L_copy_32_bytes); | |
| 1899 | |
| 1900 // Copy trailing qwords | |
| 1901 __ BIND(L_copy_8_bytes); | |
| 1902 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); | |
| 1903 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); | |
| 1904 __ incrementq(qword_count); | |
| 1905 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1906 | |
| 1907 if (is_oop) { | |
| 1908 __ jmp(L_exit); | |
| 1909 } else { | |
| 1910 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); | |
| 1911 restore_arg_regs(); | |
| 1912 __ xorq(rax, rax); // return 0 | |
| 1913 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1914 __ ret(0); | |
| 1915 } | |
| 1916 | |
| 1917 // Copy 64-byte chunks | |
| 1918 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 1919 | |
| 1920 if (is_oop) { | |
| 1921 __ BIND(L_exit); | |
| 1922 gen_write_ref_array_post_barrier(saved_to, end_to, rax); | |
| 1923 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); | |
| 1924 } else { | |
| 1925 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); | |
| 1926 } | |
| 1927 restore_arg_regs(); | |
| 1928 __ xorq(rax, rax); // return 0 | |
| 1929 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 1930 __ ret(0); | |
| 1931 | |
| 1932 return start; | |
| 1933 } | |
| 1934 | |
| 1935 // Arguments: | |
| 1936 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes | |
| 1937 // ignored | |
| 1938 // is_oop - true => oop array, so generate store check code | |
| 1939 // name - stub name string | |
| 1940 // | |
| 1941 // Inputs: | |
| 1942 // c_rarg0 - source array address | |
| 1943 // c_rarg1 - destination array address | |
| 1944 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 1945 // | |
| 1946 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) { | |
| 1947 __ align(CodeEntryAlignment); | |
| 1948 StubCodeMark mark(this, "StubRoutines", name); | |
| 1949 address start = __ pc(); | |
| 1950 | |
| 1951 Label L_copy_32_bytes, L_copy_8_bytes, L_exit; | |
| 1952 const Register from = rdi; // source array address | |
| 1953 const Register to = rsi; // destination array address | |
| 1954 const Register qword_count = rdx; // elements count | |
| 1955 const Register saved_count = rcx; | |
| 1956 | |
| 1957 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 1958 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. | |
| 1959 | |
| 1960 address disjoint_copy_entry = NULL; | |
| 1961 if (is_oop) { | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1962 assert(!UseCompressedOops, "shouldn't be called for compressed oops"); |
| 0 | 1963 disjoint_copy_entry = disjoint_oop_copy_entry; |
| 1964 oop_copy_entry = __ pc(); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1965 array_overlap_test(disjoint_oop_copy_entry, Address::times_8); |
| 0 | 1966 } else { |
| 1967 disjoint_copy_entry = disjoint_long_copy_entry; | |
| 1968 long_copy_entry = __ pc(); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
1969 array_overlap_test(disjoint_long_copy_entry, Address::times_8); |
| 0 | 1970 } |
| 1971 BLOCK_COMMENT("Entry:"); | |
| 1972 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
| 1973 | |
| 1974 array_overlap_test(disjoint_copy_entry, Address::times_8); | |
| 1975 setup_arg_regs(); // from => rdi, to => rsi, count => rdx | |
| 1976 // r9 and r10 may be used to save non-volatile registers | |
| 1977 | |
| 1978 // 'from', 'to' and 'qword_count' are now valid | |
| 1979 | |
| 1980 if (is_oop) { | |
| 1981 // Save to and count for store barrier | |
| 1982 __ movq(saved_count, qword_count); | |
| 1983 // No registers are destroyed by this call | |
| 1984 gen_write_ref_array_pre_barrier(to, saved_count); | |
| 1985 } | |
| 1986 | |
| 1987 __ jmp(L_copy_32_bytes); | |
| 1988 | |
| 1989 // Copy trailing qwords | |
| 1990 __ BIND(L_copy_8_bytes); | |
| 1991 __ movq(rax, Address(from, qword_count, Address::times_8, -8)); | |
| 1992 __ movq(Address(to, qword_count, Address::times_8, -8), rax); | |
| 1993 __ decrementq(qword_count); | |
| 1994 __ jcc(Assembler::notZero, L_copy_8_bytes); | |
| 1995 | |
| 1996 if (is_oop) { | |
| 1997 __ jmp(L_exit); | |
| 1998 } else { | |
| 1999 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); | |
| 2000 restore_arg_regs(); | |
| 2001 __ xorq(rax, rax); // return 0 | |
| 2002 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 2003 __ ret(0); | |
| 2004 } | |
| 2005 | |
| 2006 // Copy in 32-bytes chunks | |
| 2007 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); | |
| 2008 | |
| 2009 if (is_oop) { | |
| 2010 __ BIND(L_exit); | |
| 2011 __ leaq(rcx, Address(to, saved_count, Address::times_8, -8)); | |
| 2012 gen_write_ref_array_post_barrier(to, rcx, rax); | |
| 2013 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); | |
| 2014 } else { | |
| 2015 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); | |
| 2016 } | |
| 2017 restore_arg_regs(); | |
| 2018 __ xorq(rax, rax); // return 0 | |
| 2019 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 2020 __ ret(0); | |
| 2021 | |
| 2022 return start; | |
| 2023 } | |
| 2024 | |
| 2025 | |
| 2026 // Helper for generating a dynamic type check. | |
| 2027 // Smashes no registers. | |
| 2028 void generate_type_check(Register sub_klass, | |
| 2029 Register super_check_offset, | |
| 2030 Register super_klass, | |
| 2031 Label& L_success) { | |
| 2032 assert_different_registers(sub_klass, super_check_offset, super_klass); | |
| 2033 | |
| 2034 BLOCK_COMMENT("type_check:"); | |
| 2035 | |
| 2036 Label L_miss; | |
| 2037 | |
| 2038 // a couple of useful fields in sub_klass: | |
| 2039 int ss_offset = (klassOopDesc::header_size() * HeapWordSize + | |
| 2040 Klass::secondary_supers_offset_in_bytes()); | |
| 2041 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + | |
| 2042 Klass::secondary_super_cache_offset_in_bytes()); | |
| 2043 Address secondary_supers_addr(sub_klass, ss_offset); | |
| 2044 Address super_cache_addr( sub_klass, sc_offset); | |
| 2045 | |
| 2046 // if the pointers are equal, we are done (e.g., String[] elements) | |
| 2047 __ cmpq(super_klass, sub_klass); | |
| 2048 __ jcc(Assembler::equal, L_success); | |
| 2049 | |
| 2050 // check the supertype display: | |
| 2051 Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0); | |
| 2052 __ cmpq(super_klass, super_check_addr); // test the super type | |
| 2053 __ jcc(Assembler::equal, L_success); | |
| 2054 | |
| 2055 // if it was a primary super, we can just fail immediately | |
| 2056 __ cmpl(super_check_offset, sc_offset); | |
| 2057 __ jcc(Assembler::notEqual, L_miss); | |
| 2058 | |
| 2059 // Now do a linear scan of the secondary super-klass chain. | |
| 2060 // The repne_scan instruction uses fixed registers, which we must spill. | |
| 2061 // (We need a couple more temps in any case.) | |
| 2062 // This code is rarely used, so simplicity is a virtue here. | |
| 2063 inc_counter_np(SharedRuntime::_partial_subtype_ctr); | |
| 2064 { | |
| 2065 __ pushq(rax); | |
| 2066 __ pushq(rcx); | |
| 2067 __ pushq(rdi); | |
| 2068 assert_different_registers(sub_klass, super_klass, rax, rcx, rdi); | |
| 2069 | |
| 2070 __ movq(rdi, secondary_supers_addr); | |
| 2071 // Load the array length. | |
| 2072 __ movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes())); | |
| 2073 // Skip to start of data. | |
| 2074 __ addq(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); | |
| 2075 // Scan rcx words at [rdi] for occurance of rax | |
| 2076 // Set NZ/Z based on last compare | |
| 2077 __ movq(rax, super_klass); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2078 if (UseCompressedOops) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2079 // 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
|
2080 // 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
|
2081 __ encode_heap_oop(rax); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2082 __ repne_scanl(); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2083 } else { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2084 __ repne_scanq(); |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2085 } |
| 0 | 2086 |
| 2087 // Unspill the temp. registers: | |
| 2088 __ popq(rdi); | |
| 2089 __ popq(rcx); | |
| 2090 __ popq(rax); | |
| 2091 | |
| 2092 __ jcc(Assembler::notEqual, L_miss); | |
| 2093 } | |
| 2094 | |
| 2095 // Success. Cache the super we found and proceed in triumph. | |
| 2096 __ movq(super_cache_addr, super_klass); // note: rax is dead | |
| 2097 __ jmp(L_success); | |
| 2098 | |
| 2099 // Fall through on failure! | |
| 2100 __ BIND(L_miss); | |
| 2101 } | |
| 2102 | |
| 2103 // | |
| 2104 // Generate checkcasting array copy stub | |
| 2105 // | |
| 2106 // Input: | |
| 2107 // c_rarg0 - source array address | |
| 2108 // c_rarg1 - destination array address | |
| 2109 // c_rarg2 - element count, treated as ssize_t, can be zero | |
| 2110 // c_rarg3 - size_t ckoff (super_check_offset) | |
| 2111 // not Win64 | |
| 2112 // c_rarg4 - oop ckval (super_klass) | |
| 2113 // Win64 | |
| 2114 // rsp+40 - oop ckval (super_klass) | |
| 2115 // | |
| 2116 // Output: | |
| 2117 // rax == 0 - success | |
| 2118 // rax == -1^K - failure, where K is partial transfer count | |
| 2119 // | |
| 2120 address generate_checkcast_copy(const char *name) { | |
| 2121 | |
| 2122 Label L_load_element, L_store_element, L_do_card_marks, L_done; | |
| 2123 | |
| 2124 // Input registers (after setup_arg_regs) | |
| 2125 const Register from = rdi; // source array address | |
| 2126 const Register to = rsi; // destination array address | |
| 2127 const Register length = rdx; // elements count | |
| 2128 const Register ckoff = rcx; // super_check_offset | |
| 2129 const Register ckval = r8; // super_klass | |
| 2130 | |
| 2131 // Registers used as temps (r13, r14 are save-on-entry) | |
| 2132 const Register end_from = from; // source array end address | |
| 2133 const Register end_to = r13; // destination array end address | |
| 2134 const Register count = rdx; // -(count_remaining) | |
| 2135 const Register r14_length = r14; // saved copy of length | |
| 2136 // End pointers are inclusive, and if length is not zero they point | |
| 2137 // to the last unit copied: end_to[0] := end_from[0] | |
| 2138 | |
| 2139 const Register rax_oop = rax; // actual oop copied | |
| 2140 const Register r11_klass = r11; // oop._klass | |
| 2141 | |
| 2142 //--------------------------------------------------------------- | |
| 2143 // Assembler stub will be used for this call to arraycopy | |
| 2144 // if the two arrays are subtypes of Object[] but the | |
| 2145 // destination array type is not equal to or a supertype | |
| 2146 // of the source type. Each element must be separately | |
| 2147 // checked. | |
| 2148 | |
| 2149 __ align(CodeEntryAlignment); | |
| 2150 StubCodeMark mark(this, "StubRoutines", name); | |
| 2151 address start = __ pc(); | |
| 2152 | |
| 2153 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 2154 | |
| 2155 checkcast_copy_entry = __ pc(); | |
| 2156 BLOCK_COMMENT("Entry:"); | |
| 2157 | |
| 2158 #ifdef ASSERT | |
| 2159 // caller guarantees that the arrays really are different | |
| 2160 // otherwise, we would have to make conjoint checks | |
| 2161 { Label L; | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2162 array_overlap_test(L, TIMES_OOP); |
| 0 | 2163 __ stop("checkcast_copy within a single array"); |
| 2164 __ bind(L); | |
| 2165 } | |
| 2166 #endif //ASSERT | |
| 2167 | |
| 2168 // allocate spill slots for r13, r14 | |
| 2169 enum { | |
| 2170 saved_r13_offset, | |
| 2171 saved_r14_offset, | |
| 2172 saved_rbp_offset, | |
| 2173 saved_rip_offset, | |
| 2174 saved_rarg0_offset | |
| 2175 }; | |
| 2176 __ subq(rsp, saved_rbp_offset * wordSize); | |
| 2177 __ movq(Address(rsp, saved_r13_offset * wordSize), r13); | |
| 2178 __ movq(Address(rsp, saved_r14_offset * wordSize), r14); | |
| 2179 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx | |
| 2180 // ckoff => rcx, ckval => r8 | |
| 2181 // r9 and r10 may be used to save non-volatile registers | |
| 2182 #ifdef _WIN64 | |
| 2183 // last argument (#4) is on stack on Win64 | |
| 2184 const int ckval_offset = saved_rarg0_offset + 4; | |
| 2185 __ movq(ckval, Address(rsp, ckval_offset * wordSize)); | |
| 2186 #endif | |
| 2187 | |
| 2188 // check that int operands are properly extended to size_t | |
| 2189 assert_clean_int(length, rax); | |
| 2190 assert_clean_int(ckoff, rax); | |
| 2191 | |
| 2192 #ifdef ASSERT | |
| 2193 BLOCK_COMMENT("assert consistent ckoff/ckval"); | |
| 2194 // The ckoff and ckval must be mutually consistent, | |
| 2195 // even though caller generates both. | |
| 2196 { Label L; | |
| 2197 int sco_offset = (klassOopDesc::header_size() * HeapWordSize + | |
| 2198 Klass::super_check_offset_offset_in_bytes()); | |
| 2199 __ cmpl(ckoff, Address(ckval, sco_offset)); | |
| 2200 __ jcc(Assembler::equal, L); | |
| 2201 __ stop("super_check_offset inconsistent"); | |
| 2202 __ bind(L); | |
| 2203 } | |
| 2204 #endif //ASSERT | |
| 2205 | |
| 2206 // 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
|
2207 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
|
2208 Address end_to_addr(to, length, TIMES_OOP, 0); |
| 0 | 2209 // 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
|
2210 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
|
2211 Address to_element_addr(end_to, count, TIMES_OOP, 0); |
| 0 | 2212 |
| 2213 gen_write_ref_array_pre_barrier(to, count); | |
| 2214 | |
| 2215 // Copy from low to high addresses, indexed from the end of each array. | |
| 2216 __ leaq(end_from, end_from_addr); | |
| 2217 __ leaq(end_to, end_to_addr); | |
| 2218 __ movq(r14_length, length); // save a copy of the length | |
| 2219 assert(length == count, ""); // else fix next line: | |
| 2220 __ negq(count); // negate and test the length | |
| 2221 __ jcc(Assembler::notZero, L_load_element); | |
| 2222 | |
| 2223 // Empty array: Nothing to do. | |
| 2224 __ xorq(rax, rax); // return 0 on (trivial) success | |
| 2225 __ jmp(L_done); | |
| 2226 | |
| 2227 // ======== begin loop ======== | |
| 2228 // (Loop is rotated; its entry is L_load_element.) | |
| 2229 // Loop control: | |
| 2230 // for (count = -count; count != 0; count++) | |
| 2231 // Base pointers src, dst are biased by 8*(count-1),to last element. | |
| 2232 __ align(16); | |
| 2233 | |
| 2234 __ BIND(L_store_element); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2235 __ store_heap_oop(to_element_addr, rax_oop); // store the oop |
| 0 | 2236 __ incrementq(count); // increment the count toward zero |
| 2237 __ jcc(Assembler::zero, L_do_card_marks); | |
| 2238 | |
| 2239 // ======== loop entry is here ======== | |
| 2240 __ BIND(L_load_element); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2241 __ load_heap_oop(rax_oop, from_element_addr); // load the oop |
| 0 | 2242 __ testq(rax_oop, rax_oop); |
| 2243 __ jcc(Assembler::zero, L_store_element); | |
| 2244 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2245 __ load_klass(r11_klass, rax_oop);// query the object klass |
| 0 | 2246 generate_type_check(r11_klass, ckoff, ckval, L_store_element); |
| 2247 // ======== end loop ======== | |
| 2248 | |
| 2249 // It was a real error; we must depend on the caller to finish the job. | |
| 2250 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops. | |
| 2251 // Emit GC store barriers for the oops we have copied (r14 + rdx), | |
| 2252 // and report their number to the caller. | |
| 2253 assert_different_registers(rax, r14_length, count, to, end_to, rcx); | |
| 2254 __ leaq(end_to, to_element_addr); | |
| 2255 gen_write_ref_array_post_barrier(to, end_to, rcx); | |
| 2256 __ movq(rax, r14_length); // original oops | |
| 2257 __ addq(rax, count); // K = (original - remaining) oops | |
| 2258 __ notq(rax); // report (-1^K) to caller | |
| 2259 __ jmp(L_done); | |
| 2260 | |
| 2261 // Come here on success only. | |
| 2262 __ BIND(L_do_card_marks); | |
| 2263 __ addq(end_to, -wordSize); // make an inclusive end pointer | |
| 2264 gen_write_ref_array_post_barrier(to, end_to, rcx); | |
| 2265 __ xorq(rax, rax); // return 0 on success | |
| 2266 | |
| 2267 // Common exit point (success or failure). | |
| 2268 __ BIND(L_done); | |
| 2269 __ movq(r13, Address(rsp, saved_r13_offset * wordSize)); | |
| 2270 __ movq(r14, Address(rsp, saved_r14_offset * wordSize)); | |
| 2271 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); | |
| 2272 restore_arg_regs(); | |
| 2273 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 2274 __ ret(0); | |
| 2275 | |
| 2276 return start; | |
| 2277 } | |
| 2278 | |
| 2279 // | |
| 2280 // Generate 'unsafe' array copy stub | |
| 2281 // Though just as safe as the other stubs, it takes an unscaled | |
| 2282 // size_t argument instead of an element count. | |
| 2283 // | |
| 2284 // Input: | |
| 2285 // c_rarg0 - source array address | |
| 2286 // c_rarg1 - destination array address | |
| 2287 // c_rarg2 - byte count, treated as ssize_t, can be zero | |
| 2288 // | |
| 2289 // Examines the alignment of the operands and dispatches | |
| 2290 // to a long, int, short, or byte copy loop. | |
| 2291 // | |
| 2292 address generate_unsafe_copy(const char *name) { | |
| 2293 | |
| 2294 Label L_long_aligned, L_int_aligned, L_short_aligned; | |
| 2295 | |
| 2296 // Input registers (before setup_arg_regs) | |
| 2297 const Register from = c_rarg0; // source array address | |
| 2298 const Register to = c_rarg1; // destination array address | |
| 2299 const Register size = c_rarg2; // byte count (size_t) | |
| 2300 | |
| 2301 // Register used as a temp | |
| 2302 const Register bits = rax; // test copy of low bits | |
| 2303 | |
| 2304 __ align(CodeEntryAlignment); | |
| 2305 StubCodeMark mark(this, "StubRoutines", name); | |
| 2306 address start = __ pc(); | |
| 2307 | |
| 2308 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 2309 | |
| 2310 // bump this on entry, not on exit: | |
| 2311 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr); | |
| 2312 | |
| 2313 __ movq(bits, from); | |
| 2314 __ orq(bits, to); | |
| 2315 __ orq(bits, size); | |
| 2316 | |
| 2317 __ testb(bits, BytesPerLong-1); | |
| 2318 __ jccb(Assembler::zero, L_long_aligned); | |
| 2319 | |
| 2320 __ testb(bits, BytesPerInt-1); | |
| 2321 __ jccb(Assembler::zero, L_int_aligned); | |
| 2322 | |
| 2323 __ testb(bits, BytesPerShort-1); | |
| 2324 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry)); | |
| 2325 | |
| 2326 __ BIND(L_short_aligned); | |
| 2327 __ shrq(size, LogBytesPerShort); // size => short_count | |
| 2328 __ jump(RuntimeAddress(short_copy_entry)); | |
| 2329 | |
| 2330 __ BIND(L_int_aligned); | |
| 2331 __ shrq(size, LogBytesPerInt); // size => int_count | |
| 2332 __ jump(RuntimeAddress(int_copy_entry)); | |
| 2333 | |
| 2334 __ BIND(L_long_aligned); | |
| 2335 __ shrq(size, LogBytesPerLong); // size => qword_count | |
| 2336 __ jump(RuntimeAddress(long_copy_entry)); | |
| 2337 | |
| 2338 return start; | |
| 2339 } | |
| 2340 | |
| 2341 // Perform range checks on the proposed arraycopy. | |
| 2342 // Kills temp, but nothing else. | |
| 2343 // Also, clean the sign bits of src_pos and dst_pos. | |
| 2344 void arraycopy_range_checks(Register src, // source array oop (c_rarg0) | |
| 2345 Register src_pos, // source position (c_rarg1) | |
| 2346 Register dst, // destination array oo (c_rarg2) | |
| 2347 Register dst_pos, // destination position (c_rarg3) | |
| 2348 Register length, | |
| 2349 Register temp, | |
| 2350 Label& L_failed) { | |
| 2351 BLOCK_COMMENT("arraycopy_range_checks:"); | |
| 2352 | |
| 2353 // if (src_pos + length > arrayOop(src)->length()) FAIL; | |
| 2354 __ movl(temp, length); | |
| 2355 __ addl(temp, src_pos); // src_pos + length | |
| 2356 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes())); | |
| 2357 __ jcc(Assembler::above, L_failed); | |
| 2358 | |
| 2359 // if (dst_pos + length > arrayOop(dst)->length()) FAIL; | |
| 2360 __ movl(temp, length); | |
| 2361 __ addl(temp, dst_pos); // dst_pos + length | |
| 2362 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes())); | |
| 2363 __ jcc(Assembler::above, L_failed); | |
| 2364 | |
| 2365 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'. | |
| 2366 // Move with sign extension can be used since they are positive. | |
| 2367 __ movslq(src_pos, src_pos); | |
| 2368 __ movslq(dst_pos, dst_pos); | |
| 2369 | |
| 2370 BLOCK_COMMENT("arraycopy_range_checks done"); | |
| 2371 } | |
| 2372 | |
| 2373 // | |
| 2374 // Generate generic array copy stubs | |
| 2375 // | |
| 2376 // Input: | |
| 2377 // c_rarg0 - src oop | |
| 2378 // c_rarg1 - src_pos (32-bits) | |
| 2379 // c_rarg2 - dst oop | |
| 2380 // c_rarg3 - dst_pos (32-bits) | |
| 2381 // not Win64 | |
| 2382 // c_rarg4 - element count (32-bits) | |
| 2383 // Win64 | |
| 2384 // rsp+40 - element count (32-bits) | |
| 2385 // | |
| 2386 // Output: | |
| 2387 // rax == 0 - success | |
| 2388 // rax == -1^K - failure, where K is partial transfer count | |
| 2389 // | |
| 2390 address generate_generic_copy(const char *name) { | |
| 2391 | |
| 2392 Label L_failed, L_failed_0, L_objArray; | |
| 2393 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs; | |
| 2394 | |
| 2395 // Input registers | |
| 2396 const Register src = c_rarg0; // source array oop | |
| 2397 const Register src_pos = c_rarg1; // source position | |
| 2398 const Register dst = c_rarg2; // destination array oop | |
| 2399 const Register dst_pos = c_rarg3; // destination position | |
| 2400 // elements count is on stack on Win64 | |
| 2401 #ifdef _WIN64 | |
| 2402 #define C_RARG4 Address(rsp, 6 * wordSize) | |
| 2403 #else | |
| 2404 #define C_RARG4 c_rarg4 | |
| 2405 #endif | |
| 2406 | |
| 2407 { int modulus = CodeEntryAlignment; | |
| 2408 int target = modulus - 5; // 5 = sizeof jmp(L_failed) | |
| 2409 int advance = target - (__ offset() % modulus); | |
| 2410 if (advance < 0) advance += modulus; | |
| 2411 if (advance > 0) __ nop(advance); | |
| 2412 } | |
| 2413 StubCodeMark mark(this, "StubRoutines", name); | |
| 2414 | |
| 2415 // Short-hop target to L_failed. Makes for denser prologue code. | |
| 2416 __ BIND(L_failed_0); | |
| 2417 __ jmp(L_failed); | |
| 2418 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed"); | |
| 2419 | |
| 2420 __ align(CodeEntryAlignment); | |
| 2421 address start = __ pc(); | |
| 2422 | |
| 2423 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 2424 | |
| 2425 // bump this on entry, not on exit: | |
| 2426 inc_counter_np(SharedRuntime::_generic_array_copy_ctr); | |
| 2427 | |
| 2428 //----------------------------------------------------------------------- | |
| 2429 // Assembler stub will be used for this call to arraycopy | |
| 2430 // if the following conditions are met: | |
| 2431 // | |
| 2432 // (1) src and dst must not be null. | |
| 2433 // (2) src_pos must not be negative. | |
| 2434 // (3) dst_pos must not be negative. | |
| 2435 // (4) length must not be negative. | |
| 2436 // (5) src klass and dst klass should be the same and not NULL. | |
| 2437 // (6) src and dst should be arrays. | |
| 2438 // (7) src_pos + length must not exceed length of src. | |
| 2439 // (8) dst_pos + length must not exceed length of dst. | |
| 2440 // | |
| 2441 | |
| 2442 // if (src == NULL) return -1; | |
| 2443 __ testq(src, src); // src oop | |
| 2444 size_t j1off = __ offset(); | |
| 2445 __ jccb(Assembler::zero, L_failed_0); | |
| 2446 | |
| 2447 // if (src_pos < 0) return -1; | |
| 2448 __ testl(src_pos, src_pos); // src_pos (32-bits) | |
| 2449 __ jccb(Assembler::negative, L_failed_0); | |
| 2450 | |
| 2451 // if (dst == NULL) return -1; | |
| 2452 __ testq(dst, dst); // dst oop | |
| 2453 __ jccb(Assembler::zero, L_failed_0); | |
| 2454 | |
| 2455 // if (dst_pos < 0) return -1; | |
| 2456 __ testl(dst_pos, dst_pos); // dst_pos (32-bits) | |
| 2457 size_t j4off = __ offset(); | |
| 2458 __ jccb(Assembler::negative, L_failed_0); | |
| 2459 | |
| 2460 // The first four tests are very dense code, | |
| 2461 // but not quite dense enough to put four | |
| 2462 // jumps in a 16-byte instruction fetch buffer. | |
| 2463 // That's good, because some branch predicters | |
| 2464 // do not like jumps so close together. | |
| 2465 // Make sure of this. | |
| 2466 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps"); | |
| 2467 | |
| 2468 // registers used as temp | |
| 2469 const Register r11_length = r11; // elements count to copy | |
| 2470 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
|
2471 const Register r9_dst_klass = r9; // dest array klass |
| 0 | 2472 |
| 2473 // if (length < 0) return -1; | |
| 2474 __ movl(r11_length, C_RARG4); // length (elements count, 32-bits value) | |
| 2475 __ testl(r11_length, r11_length); | |
| 2476 __ jccb(Assembler::negative, L_failed_0); | |
| 2477 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2478 __ load_klass(r10_src_klass, src); |
| 0 | 2479 #ifdef ASSERT |
| 2480 // assert(src->klass() != NULL); | |
| 2481 BLOCK_COMMENT("assert klasses not null"); | |
| 2482 { Label L1, L2; | |
| 2483 __ testq(r10_src_klass, r10_src_klass); | |
| 2484 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL | |
| 2485 __ bind(L1); | |
| 2486 __ stop("broken null klass"); | |
| 2487 __ bind(L2); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2488 __ 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
|
2489 __ cmpq(r9_dst_klass, 0); |
| 0 | 2490 __ jcc(Assembler::equal, L1); // this would be broken also |
| 2491 BLOCK_COMMENT("assert done"); | |
| 2492 } | |
| 2493 #endif | |
| 2494 | |
| 2495 // Load layout helper (32-bits) | |
| 2496 // | |
| 2497 // |array_tag| | header_size | element_type | |log2_element_size| | |
| 2498 // 32 30 24 16 8 2 0 | |
| 2499 // | |
| 2500 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 | |
| 2501 // | |
| 2502 | |
| 2503 int lh_offset = klassOopDesc::header_size() * HeapWordSize + | |
| 2504 Klass::layout_helper_offset_in_bytes(); | |
| 2505 | |
| 2506 const Register rax_lh = rax; // layout helper | |
| 2507 | |
| 2508 __ movl(rax_lh, Address(r10_src_klass, lh_offset)); | |
| 2509 | |
| 2510 // Handle objArrays completely differently... | |
| 2511 jint objArray_lh = Klass::array_layout_helper(T_OBJECT); | |
| 2512 __ cmpl(rax_lh, objArray_lh); | |
| 2513 __ jcc(Assembler::equal, L_objArray); | |
| 2514 | |
| 2515 // 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
|
2516 __ 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
|
2517 __ cmpq(r10_src_klass, r9_dst_klass); |
| 0 | 2518 __ jcc(Assembler::notEqual, L_failed); |
| 2519 | |
| 2520 // if (!src->is_Array()) return -1; | |
| 2521 __ cmpl(rax_lh, Klass::_lh_neutral_value); | |
| 2522 __ jcc(Assembler::greaterEqual, L_failed); | |
| 2523 | |
| 2524 // At this point, it is known to be a typeArray (array_tag 0x3). | |
| 2525 #ifdef ASSERT | |
| 2526 { Label L; | |
| 2527 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift)); | |
| 2528 __ jcc(Assembler::greaterEqual, L); | |
| 2529 __ stop("must be a primitive array"); | |
| 2530 __ bind(L); | |
| 2531 } | |
| 2532 #endif | |
| 2533 | |
| 2534 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, | |
| 2535 r10, L_failed); | |
| 2536 | |
| 2537 // typeArrayKlass | |
| 2538 // | |
| 2539 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); | |
| 2540 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); | |
| 2541 // | |
| 2542 | |
| 2543 const Register r10_offset = r10; // array offset | |
| 2544 const Register rax_elsize = rax_lh; // element size | |
| 2545 | |
| 2546 __ movl(r10_offset, rax_lh); | |
| 2547 __ shrl(r10_offset, Klass::_lh_header_size_shift); | |
| 2548 __ andq(r10_offset, Klass::_lh_header_size_mask); // array_offset | |
| 2549 __ addq(src, r10_offset); // src array offset | |
| 2550 __ addq(dst, r10_offset); // dst array offset | |
| 2551 BLOCK_COMMENT("choose copy loop based on element size"); | |
| 2552 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize | |
| 2553 | |
| 2554 // next registers should be set before the jump to corresponding stub | |
| 2555 const Register from = c_rarg0; // source array address | |
| 2556 const Register to = c_rarg1; // destination array address | |
| 2557 const Register count = c_rarg2; // elements count | |
| 2558 | |
| 2559 // 'from', 'to', 'count' registers should be set in such order | |
| 2560 // since they are the same as 'src', 'src_pos', 'dst'. | |
| 2561 | |
| 2562 __ BIND(L_copy_bytes); | |
| 2563 __ cmpl(rax_elsize, 0); | |
| 2564 __ jccb(Assembler::notEqual, L_copy_shorts); | |
| 2565 __ leaq(from, Address(src, src_pos, Address::times_1, 0));// src_addr | |
| 2566 __ leaq(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr | |
| 2567 __ movslq(count, r11_length); // length | |
| 2568 __ jump(RuntimeAddress(byte_copy_entry)); | |
| 2569 | |
| 2570 __ BIND(L_copy_shorts); | |
| 2571 __ cmpl(rax_elsize, LogBytesPerShort); | |
| 2572 __ jccb(Assembler::notEqual, L_copy_ints); | |
| 2573 __ leaq(from, Address(src, src_pos, Address::times_2, 0));// src_addr | |
| 2574 __ leaq(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr | |
| 2575 __ movslq(count, r11_length); // length | |
| 2576 __ jump(RuntimeAddress(short_copy_entry)); | |
| 2577 | |
| 2578 __ BIND(L_copy_ints); | |
| 2579 __ cmpl(rax_elsize, LogBytesPerInt); | |
| 2580 __ jccb(Assembler::notEqual, L_copy_longs); | |
| 2581 __ leaq(from, Address(src, src_pos, Address::times_4, 0));// src_addr | |
| 2582 __ leaq(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr | |
| 2583 __ movslq(count, r11_length); // length | |
| 2584 __ jump(RuntimeAddress(int_copy_entry)); | |
| 2585 | |
| 2586 __ BIND(L_copy_longs); | |
| 2587 #ifdef ASSERT | |
| 2588 { Label L; | |
| 2589 __ cmpl(rax_elsize, LogBytesPerLong); | |
| 2590 __ jcc(Assembler::equal, L); | |
| 2591 __ stop("must be long copy, but elsize is wrong"); | |
| 2592 __ bind(L); | |
| 2593 } | |
| 2594 #endif | |
| 2595 __ leaq(from, Address(src, src_pos, Address::times_8, 0));// src_addr | |
| 2596 __ leaq(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr | |
| 2597 __ movslq(count, r11_length); // length | |
| 2598 __ jump(RuntimeAddress(long_copy_entry)); | |
| 2599 | |
| 2600 // objArrayKlass | |
| 2601 __ BIND(L_objArray); | |
| 2602 // live at this point: r10_src_klass, src[_pos], dst[_pos] | |
| 2603 | |
| 2604 Label L_plain_copy, L_checkcast_copy; | |
| 2605 // 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
|
2606 __ 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
|
2607 __ cmpq(r10_src_klass, r9_dst_klass); // usual case is exact equality |
| 0 | 2608 __ jcc(Assembler::notEqual, L_checkcast_copy); |
| 2609 | |
| 2610 // Identically typed arrays can be copied without element-wise checks. | |
| 2611 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, | |
| 2612 r10, L_failed); | |
| 2613 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2614 __ leaq(from, Address(src, src_pos, TIMES_OOP, |
| 0 | 2615 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
|
2616 __ 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
|
2617 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr |
| 0 | 2618 __ movslq(count, r11_length); // length |
| 2619 __ BIND(L_plain_copy); | |
| 2620 __ jump(RuntimeAddress(oop_copy_entry)); | |
| 2621 | |
| 2622 __ BIND(L_checkcast_copy); | |
| 2623 // live at this point: r10_src_klass, !r11_length | |
| 2624 { | |
| 2625 // assert(r11_length == C_RARG4); // will reload from here | |
| 2626 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
|
2627 __ load_klass(r11_dst_klass, dst); |
| 0 | 2628 |
| 2629 // Before looking at dst.length, make sure dst is also an objArray. | |
| 2630 __ cmpl(Address(r11_dst_klass, lh_offset), objArray_lh); | |
| 2631 __ jcc(Assembler::notEqual, L_failed); | |
| 2632 | |
| 2633 // It is safe to examine both src.length and dst.length. | |
| 2634 #ifndef _WIN64 | |
| 2635 arraycopy_range_checks(src, src_pos, dst, dst_pos, C_RARG4, | |
| 2636 rax, L_failed); | |
| 2637 #else | |
| 2638 __ movl(r11_length, C_RARG4); // reload | |
| 2639 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, | |
| 2640 rax, L_failed); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2641 __ load_klass(r11_dst_klass, dst); // reload |
| 0 | 2642 #endif |
| 2643 | |
| 2644 // 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
|
2645 __ leaq(from, Address(src, src_pos, TIMES_OOP, |
| 0 | 2646 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
|
2647 __ leaq(to, Address(dst, dst_pos, TIMES_OOP, |
| 0 | 2648 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); |
| 2649 __ movl(count, C_RARG4); // length (reloaded) | |
| 2650 Register sco_temp = c_rarg3; // this register is free now | |
| 2651 assert_different_registers(from, to, count, sco_temp, | |
| 2652 r11_dst_klass, r10_src_klass); | |
| 2653 assert_clean_int(count, sco_temp); | |
| 2654 | |
| 2655 // Generate the type check. | |
| 2656 int sco_offset = (klassOopDesc::header_size() * HeapWordSize + | |
| 2657 Klass::super_check_offset_offset_in_bytes()); | |
| 2658 __ movl(sco_temp, Address(r11_dst_klass, sco_offset)); | |
| 2659 assert_clean_int(sco_temp, rax); | |
| 2660 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy); | |
| 2661 | |
| 2662 // Fetch destination element klass from the objArrayKlass header. | |
| 2663 int ek_offset = (klassOopDesc::header_size() * HeapWordSize + | |
| 2664 objArrayKlass::element_klass_offset_in_bytes()); | |
| 2665 __ movq(r11_dst_klass, Address(r11_dst_klass, ek_offset)); | |
| 2666 __ movl(sco_temp, Address(r11_dst_klass, sco_offset)); | |
| 2667 assert_clean_int(sco_temp, rax); | |
| 2668 | |
| 2669 // the checkcast_copy loop needs two extra arguments: | |
| 2670 assert(c_rarg3 == sco_temp, "#3 already in place"); | |
| 2671 __ movq(C_RARG4, r11_dst_klass); // dst.klass.element_klass | |
| 2672 __ jump(RuntimeAddress(checkcast_copy_entry)); | |
| 2673 } | |
| 2674 | |
| 2675 __ BIND(L_failed); | |
| 2676 __ xorq(rax, rax); | |
| 2677 __ notq(rax); // return -1 | |
| 2678 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 2679 __ ret(0); | |
| 2680 | |
| 2681 return start; | |
| 2682 } | |
| 2683 | |
| 2684 #undef length_arg | |
| 2685 | |
| 2686 void generate_arraycopy_stubs() { | |
| 2687 // Call the conjoint generation methods immediately after | |
| 2688 // the disjoint ones so that short branches from the former | |
| 2689 // to the latter can be generated. | |
| 2690 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy"); | |
| 2691 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, "jbyte_arraycopy"); | |
| 2692 | |
| 2693 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy"); | |
| 2694 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, "jshort_arraycopy"); | |
| 2695 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2696 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
|
2697 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, "jint_arraycopy"); |
| 0 | 2698 |
| 2699 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, "jlong_disjoint_arraycopy"); | |
| 2700 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, "jlong_arraycopy"); | |
| 2701 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2702 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2703 if (UseCompressedOops) { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2704 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
|
2705 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
|
2706 } else { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
16
diff
changeset
|
2707 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
|
2708 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
|
2709 } |
| 0 | 2710 |
| 2711 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy"); | |
| 2712 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy"); | |
| 2713 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy"); | |
| 2714 | |
| 2715 // We don't generate specialized code for HeapWord-aligned source | |
| 2716 // arrays, so just use the code we've already generated | |
| 2717 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy; | |
| 2718 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy; | |
| 2719 | |
| 2720 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy; | |
| 2721 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy; | |
| 2722 | |
| 2723 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy; | |
| 2724 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy; | |
| 2725 | |
| 2726 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy; | |
| 2727 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy; | |
| 2728 | |
| 2729 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy; | |
| 2730 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; | |
| 2731 } | |
| 2732 | |
| 2733 #undef __ | |
| 2734 #define __ masm-> | |
| 2735 | |
| 2736 // Continuation point for throwing of implicit exceptions that are | |
| 2737 // not handled in the current activation. Fabricates an exception | |
| 2738 // oop and initiates normal exception dispatching in this | |
| 2739 // frame. Since we need to preserve callee-saved values (currently | |
| 2740 // only for C2, but done for C1 as well) we need a callee-saved oop | |
| 2741 // map and therefore have to make these stubs into RuntimeStubs | |
| 2742 // rather than BufferBlobs. If the compiler needs all registers to | |
| 2743 // be preserved between the fault point and the exception handler | |
| 2744 // then it must assume responsibility for that in | |
| 2745 // AbstractCompiler::continuation_for_implicit_null_exception or | |
| 2746 // continuation_for_implicit_division_by_zero_exception. All other | |
| 2747 // implicit exceptions (e.g., NullPointerException or | |
| 2748 // AbstractMethodError on entry) are either at call sites or | |
| 2749 // otherwise assume that stack unwinding will be initiated, so | |
| 2750 // caller saved registers were assumed volatile in the compiler. | |
| 2751 address generate_throw_exception(const char* name, | |
| 2752 address runtime_entry, | |
| 2753 bool restore_saved_exception_pc) { | |
| 2754 // Information about frame layout at time of blocking runtime call. | |
| 2755 // Note that we only have to preserve callee-saved registers since | |
| 2756 // the compilers are responsible for supplying a continuation point | |
| 2757 // if they expect all registers to be preserved. | |
| 2758 enum layout { | |
| 2759 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt, | |
| 2760 rbp_off2, | |
| 2761 return_off, | |
| 2762 return_off2, | |
| 2763 framesize // inclusive of return address | |
| 2764 }; | |
| 2765 | |
| 2766 int insts_size = 512; | |
| 2767 int locs_size = 64; | |
| 2768 | |
| 2769 CodeBuffer code(name, insts_size, locs_size); | |
| 2770 OopMapSet* oop_maps = new OopMapSet(); | |
| 2771 MacroAssembler* masm = new MacroAssembler(&code); | |
| 2772 | |
| 2773 address start = __ pc(); | |
| 2774 | |
| 2775 // This is an inlined and slightly modified version of call_VM | |
| 2776 // which has the ability to fetch the return PC out of | |
| 2777 // thread-local storage and also sets up last_Java_sp slightly | |
| 2778 // differently than the real call_VM | |
| 2779 if (restore_saved_exception_pc) { | |
| 2780 __ movq(rax, | |
| 2781 Address(r15_thread, | |
| 2782 in_bytes(JavaThread::saved_exception_pc_offset()))); | |
| 2783 __ pushq(rax); | |
| 2784 } | |
| 2785 | |
| 2786 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
| 2787 | |
| 2788 assert(is_even(framesize/2), "sp not 16-byte aligned"); | |
| 2789 | |
| 2790 // return address and rbp are already in place | |
| 2791 __ subq(rsp, (framesize-4) << LogBytesPerInt); // prolog | |
| 2792 | |
| 2793 int frame_complete = __ pc() - start; | |
| 2794 | |
| 2795 // Set up last_Java_sp and last_Java_fp | |
| 2796 __ set_last_Java_frame(rsp, rbp, NULL); | |
| 2797 | |
| 2798 // Call runtime | |
| 2799 __ movq(c_rarg0, r15_thread); | |
| 2800 BLOCK_COMMENT("call runtime_entry"); | |
| 2801 __ call(RuntimeAddress(runtime_entry)); | |
| 2802 | |
| 2803 // Generate oop map | |
| 2804 OopMap* map = new OopMap(framesize, 0); | |
| 2805 | |
| 2806 oop_maps->add_gc_map(__ pc() - start, map); | |
| 2807 | |
| 2808 __ reset_last_Java_frame(true, false); | |
| 2809 | |
| 2810 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
| 2811 | |
| 2812 // check for pending exceptions | |
| 2813 #ifdef ASSERT | |
| 2814 Label L; | |
| 2815 __ cmpq(Address(r15_thread, Thread::pending_exception_offset()), | |
| 2816 (int) NULL); | |
| 2817 __ jcc(Assembler::notEqual, L); | |
| 2818 __ should_not_reach_here(); | |
| 2819 __ bind(L); | |
| 2820 #endif // ASSERT | |
| 2821 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); | |
| 2822 | |
| 2823 | |
| 2824 // codeBlob framesize is in words (not VMRegImpl::slot_size) | |
| 2825 RuntimeStub* stub = | |
| 2826 RuntimeStub::new_runtime_stub(name, | |
| 2827 &code, | |
| 2828 frame_complete, | |
| 2829 (framesize >> (LogBytesPerWord - LogBytesPerInt)), | |
| 2830 oop_maps, false); | |
| 2831 return stub->entry_point(); | |
| 2832 } | |
| 2833 | |
| 2834 // Initialization | |
| 2835 void generate_initial() { | |
| 2836 // Generates all stubs and initializes the entry points | |
| 2837 | |
| 2838 // This platform-specific stub is needed by generate_call_stub() | |
| 2839 StubRoutines::amd64::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80); | |
| 2840 | |
| 2841 // entry points that exist in all platforms Note: This is code | |
| 2842 // that could be shared among different platforms - however the | |
| 2843 // benefit seems to be smaller than the disadvantage of having a | |
| 2844 // much more complicated generator structure. See also comment in | |
| 2845 // stubRoutines.hpp. | |
| 2846 | |
| 2847 StubRoutines::_forward_exception_entry = generate_forward_exception(); | |
| 2848 | |
| 2849 StubRoutines::_call_stub_entry = | |
| 2850 generate_call_stub(StubRoutines::_call_stub_return_address); | |
| 2851 | |
| 2852 // is referenced by megamorphic call | |
| 2853 StubRoutines::_catch_exception_entry = generate_catch_exception(); | |
| 2854 | |
| 2855 // atomic calls | |
| 2856 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); | |
| 2857 StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr(); | |
| 2858 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg(); | |
| 2859 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long(); | |
| 2860 StubRoutines::_atomic_add_entry = generate_atomic_add(); | |
| 2861 StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr(); | |
| 2862 StubRoutines::_fence_entry = generate_orderaccess_fence(); | |
| 2863 | |
| 2864 StubRoutines::_handler_for_unsafe_access_entry = | |
| 2865 generate_handler_for_unsafe_access(); | |
| 2866 | |
| 2867 // platform dependent | |
| 2868 StubRoutines::amd64::_get_previous_fp_entry = generate_get_previous_fp(); | |
| 2869 | |
| 2870 StubRoutines::amd64::_verify_mxcsr_entry = generate_verify_mxcsr(); | |
| 2871 } | |
| 2872 | |
| 2873 void generate_all() { | |
| 2874 // Generates all stubs and initializes the entry points | |
| 2875 | |
| 2876 // These entry points require SharedInfo::stack0 to be set up in | |
| 2877 // non-core builds and need to be relocatable, so they each | |
| 2878 // fabricate a RuntimeStub internally. | |
| 2879 StubRoutines::_throw_AbstractMethodError_entry = | |
| 2880 generate_throw_exception("AbstractMethodError throw_exception", | |
| 2881 CAST_FROM_FN_PTR(address, | |
| 2882 SharedRuntime:: | |
| 2883 throw_AbstractMethodError), | |
| 2884 false); | |
| 2885 | |
|
16
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2886 StubRoutines::_throw_IncompatibleClassChangeError_entry = |
|
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2887 generate_throw_exception("IncompatibleClassChangeError throw_exception", |
|
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2888 CAST_FROM_FN_PTR(address, |
|
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2889 SharedRuntime:: |
|
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2890 throw_IncompatibleClassChangeError), |
|
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2891 false); |
|
f8236e79048a
6664627: Merge changes made only in hotspot 11 forward to jdk 7
dcubed
parents:
0
diff
changeset
|
2892 |
| 0 | 2893 StubRoutines::_throw_ArithmeticException_entry = |
| 2894 generate_throw_exception("ArithmeticException throw_exception", | |
| 2895 CAST_FROM_FN_PTR(address, | |
| 2896 SharedRuntime:: | |
| 2897 throw_ArithmeticException), | |
| 2898 true); | |
| 2899 | |
| 2900 StubRoutines::_throw_NullPointerException_entry = | |
| 2901 generate_throw_exception("NullPointerException throw_exception", | |
| 2902 CAST_FROM_FN_PTR(address, | |
| 2903 SharedRuntime:: | |
| 2904 throw_NullPointerException), | |
| 2905 true); | |
| 2906 | |
| 2907 StubRoutines::_throw_NullPointerException_at_call_entry = | |
| 2908 generate_throw_exception("NullPointerException at call throw_exception", | |
| 2909 CAST_FROM_FN_PTR(address, | |
| 2910 SharedRuntime:: | |
| 2911 throw_NullPointerException_at_call), | |
| 2912 false); | |
| 2913 | |
| 2914 StubRoutines::_throw_StackOverflowError_entry = | |
| 2915 generate_throw_exception("StackOverflowError throw_exception", | |
| 2916 CAST_FROM_FN_PTR(address, | |
| 2917 SharedRuntime:: | |
| 2918 throw_StackOverflowError), | |
| 2919 false); | |
| 2920 | |
| 2921 // entry points that are platform specific | |
| 2922 StubRoutines::amd64::_f2i_fixup = generate_f2i_fixup(); | |
| 2923 StubRoutines::amd64::_f2l_fixup = generate_f2l_fixup(); | |
| 2924 StubRoutines::amd64::_d2i_fixup = generate_d2i_fixup(); | |
| 2925 StubRoutines::amd64::_d2l_fixup = generate_d2l_fixup(); | |
| 2926 | |
| 2927 StubRoutines::amd64::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF); | |
| 2928 StubRoutines::amd64::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000); | |
| 2929 StubRoutines::amd64::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF); | |
| 2930 StubRoutines::amd64::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000); | |
| 2931 | |
| 2932 // support for verify_oop (must happen after universe_init) | |
| 2933 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop(); | |
| 2934 | |
| 2935 // arraycopy stubs used by compilers | |
| 2936 generate_arraycopy_stubs(); | |
| 2937 } | |
| 2938 | |
| 2939 public: | |
| 2940 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { | |
| 2941 if (all) { | |
| 2942 generate_all(); | |
| 2943 } else { | |
| 2944 generate_initial(); | |
| 2945 } | |
| 2946 } | |
| 2947 }; // end class declaration | |
| 2948 | |
| 2949 address StubGenerator::disjoint_byte_copy_entry = NULL; | |
| 2950 address StubGenerator::disjoint_short_copy_entry = NULL; | |
| 2951 address StubGenerator::disjoint_int_copy_entry = NULL; | |
| 2952 address StubGenerator::disjoint_long_copy_entry = NULL; | |
| 2953 address StubGenerator::disjoint_oop_copy_entry = NULL; | |
| 2954 | |
| 2955 address StubGenerator::byte_copy_entry = NULL; | |
| 2956 address StubGenerator::short_copy_entry = NULL; | |
| 2957 address StubGenerator::int_copy_entry = NULL; | |
| 2958 address StubGenerator::long_copy_entry = NULL; | |
| 2959 address StubGenerator::oop_copy_entry = NULL; | |
| 2960 | |
| 2961 address StubGenerator::checkcast_copy_entry = NULL; | |
| 2962 | |
| 2963 void StubGenerator_generate(CodeBuffer* code, bool all) { | |
| 2964 StubGenerator g(code, all); | |
| 2965 } |