Fri, 18 Nov 2011 10:29:27 -0800
7110489: C1: 64-bit tiered with ForceUnreachable: assert(reachable(src)) failed: Address should be reachable
Reviewed-by: kvn, iveresov, twisti
duke@435 | 1 | /* |
iveresov@2595 | 2 | * Copyright (c) 2003, 2011, Oracle and/or its affiliates. All rights reserved. |
duke@435 | 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
duke@435 | 4 | * |
duke@435 | 5 | * This code is free software; you can redistribute it and/or modify it |
duke@435 | 6 | * under the terms of the GNU General Public License version 2 only, as |
duke@435 | 7 | * published by the Free Software Foundation. |
duke@435 | 8 | * |
duke@435 | 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
duke@435 | 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
duke@435 | 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
duke@435 | 12 | * version 2 for more details (a copy is included in the LICENSE file that |
duke@435 | 13 | * accompanied this code). |
duke@435 | 14 | * |
duke@435 | 15 | * You should have received a copy of the GNU General Public License version |
duke@435 | 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
duke@435 | 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
duke@435 | 18 | * |
trims@1907 | 19 | * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
trims@1907 | 20 | * or visit www.oracle.com if you need additional information or have any |
trims@1907 | 21 | * questions. |
duke@435 | 22 | * |
duke@435 | 23 | */ |
duke@435 | 24 | |
stefank@2314 | 25 | #include "precompiled.hpp" |
stefank@2314 | 26 | #include "asm/assembler.hpp" |
stefank@2314 | 27 | #include "assembler_x86.inline.hpp" |
stefank@2314 | 28 | #include "interpreter/interpreter.hpp" |
stefank@2314 | 29 | #include "nativeInst_x86.hpp" |
stefank@2314 | 30 | #include "oops/instanceOop.hpp" |
stefank@2314 | 31 | #include "oops/methodOop.hpp" |
stefank@2314 | 32 | #include "oops/objArrayKlass.hpp" |
stefank@2314 | 33 | #include "oops/oop.inline.hpp" |
stefank@2314 | 34 | #include "prims/methodHandles.hpp" |
stefank@2314 | 35 | #include "runtime/frame.inline.hpp" |
stefank@2314 | 36 | #include "runtime/handles.inline.hpp" |
stefank@2314 | 37 | #include "runtime/sharedRuntime.hpp" |
stefank@2314 | 38 | #include "runtime/stubCodeGenerator.hpp" |
stefank@2314 | 39 | #include "runtime/stubRoutines.hpp" |
stefank@2314 | 40 | #include "utilities/top.hpp" |
stefank@2314 | 41 | #ifdef TARGET_OS_FAMILY_linux |
stefank@2314 | 42 | # include "thread_linux.inline.hpp" |
stefank@2314 | 43 | #endif |
stefank@2314 | 44 | #ifdef TARGET_OS_FAMILY_solaris |
stefank@2314 | 45 | # include "thread_solaris.inline.hpp" |
stefank@2314 | 46 | #endif |
stefank@2314 | 47 | #ifdef TARGET_OS_FAMILY_windows |
stefank@2314 | 48 | # include "thread_windows.inline.hpp" |
stefank@2314 | 49 | #endif |
never@3156 | 50 | #ifdef TARGET_OS_FAMILY_bsd |
never@3156 | 51 | # include "thread_bsd.inline.hpp" |
never@3156 | 52 | #endif |
stefank@2314 | 53 | #ifdef COMPILER2 |
stefank@2314 | 54 | #include "opto/runtime.hpp" |
stefank@2314 | 55 | #endif |
duke@435 | 56 | |
duke@435 | 57 | // Declaration and definition of StubGenerator (no .hpp file). |
duke@435 | 58 | // For a more detailed description of the stub routine structure |
duke@435 | 59 | // see the comment in stubRoutines.hpp |
duke@435 | 60 | |
duke@435 | 61 | #define __ _masm-> |
coleenp@548 | 62 | #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8) |
never@739 | 63 | #define a__ ((Assembler*)_masm)-> |
duke@435 | 64 | |
duke@435 | 65 | #ifdef PRODUCT |
duke@435 | 66 | #define BLOCK_COMMENT(str) /* nothing */ |
duke@435 | 67 | #else |
duke@435 | 68 | #define BLOCK_COMMENT(str) __ block_comment(str) |
duke@435 | 69 | #endif |
duke@435 | 70 | |
duke@435 | 71 | #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") |
duke@435 | 72 | const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions |
duke@435 | 73 | |
duke@435 | 74 | // Stub Code definitions |
duke@435 | 75 | |
duke@435 | 76 | static address handle_unsafe_access() { |
duke@435 | 77 | JavaThread* thread = JavaThread::current(); |
duke@435 | 78 | address pc = thread->saved_exception_pc(); |
duke@435 | 79 | // pc is the instruction which we must emulate |
duke@435 | 80 | // doing a no-op is fine: return garbage from the load |
duke@435 | 81 | // therefore, compute npc |
duke@435 | 82 | address npc = Assembler::locate_next_instruction(pc); |
duke@435 | 83 | |
duke@435 | 84 | // request an async exception |
duke@435 | 85 | thread->set_pending_unsafe_access_error(); |
duke@435 | 86 | |
duke@435 | 87 | // return address of next instruction to execute |
duke@435 | 88 | return npc; |
duke@435 | 89 | } |
duke@435 | 90 | |
duke@435 | 91 | class StubGenerator: public StubCodeGenerator { |
duke@435 | 92 | private: |
duke@435 | 93 | |
duke@435 | 94 | #ifdef PRODUCT |
duke@435 | 95 | #define inc_counter_np(counter) (0) |
duke@435 | 96 | #else |
duke@435 | 97 | void inc_counter_np_(int& counter) { |
never@3314 | 98 | // This can destroy rscratch1 if counter is far from the code cache |
duke@435 | 99 | __ incrementl(ExternalAddress((address)&counter)); |
duke@435 | 100 | } |
duke@435 | 101 | #define inc_counter_np(counter) \ |
duke@435 | 102 | BLOCK_COMMENT("inc_counter " #counter); \ |
duke@435 | 103 | inc_counter_np_(counter); |
duke@435 | 104 | #endif |
duke@435 | 105 | |
duke@435 | 106 | // Call stubs are used to call Java from C |
duke@435 | 107 | // |
duke@435 | 108 | // Linux Arguments: |
duke@435 | 109 | // c_rarg0: call wrapper address address |
duke@435 | 110 | // c_rarg1: result address |
duke@435 | 111 | // c_rarg2: result type BasicType |
duke@435 | 112 | // c_rarg3: method methodOop |
duke@435 | 113 | // c_rarg4: (interpreter) entry point address |
duke@435 | 114 | // c_rarg5: parameters intptr_t* |
duke@435 | 115 | // 16(rbp): parameter size (in words) int |
duke@435 | 116 | // 24(rbp): thread Thread* |
duke@435 | 117 | // |
duke@435 | 118 | // [ return_from_Java ] <--- rsp |
duke@435 | 119 | // [ argument word n ] |
duke@435 | 120 | // ... |
duke@435 | 121 | // -12 [ argument word 1 ] |
duke@435 | 122 | // -11 [ saved r15 ] <--- rsp_after_call |
duke@435 | 123 | // -10 [ saved r14 ] |
duke@435 | 124 | // -9 [ saved r13 ] |
duke@435 | 125 | // -8 [ saved r12 ] |
duke@435 | 126 | // -7 [ saved rbx ] |
duke@435 | 127 | // -6 [ call wrapper ] |
duke@435 | 128 | // -5 [ result ] |
duke@435 | 129 | // -4 [ result type ] |
duke@435 | 130 | // -3 [ method ] |
duke@435 | 131 | // -2 [ entry point ] |
duke@435 | 132 | // -1 [ parameters ] |
duke@435 | 133 | // 0 [ saved rbp ] <--- rbp |
duke@435 | 134 | // 1 [ return address ] |
duke@435 | 135 | // 2 [ parameter size ] |
duke@435 | 136 | // 3 [ thread ] |
duke@435 | 137 | // |
duke@435 | 138 | // Windows Arguments: |
duke@435 | 139 | // c_rarg0: call wrapper address address |
duke@435 | 140 | // c_rarg1: result address |
duke@435 | 141 | // c_rarg2: result type BasicType |
duke@435 | 142 | // c_rarg3: method methodOop |
duke@435 | 143 | // 48(rbp): (interpreter) entry point address |
duke@435 | 144 | // 56(rbp): parameters intptr_t* |
duke@435 | 145 | // 64(rbp): parameter size (in words) int |
duke@435 | 146 | // 72(rbp): thread Thread* |
duke@435 | 147 | // |
duke@435 | 148 | // [ return_from_Java ] <--- rsp |
duke@435 | 149 | // [ argument word n ] |
duke@435 | 150 | // ... |
iveresov@2689 | 151 | // -28 [ argument word 1 ] |
iveresov@2689 | 152 | // -27 [ saved xmm15 ] <--- rsp_after_call |
iveresov@2689 | 153 | // [ saved xmm7-xmm14 ] |
iveresov@2689 | 154 | // -9 [ saved xmm6 ] (each xmm register takes 2 slots) |
iveresov@2689 | 155 | // -7 [ saved r15 ] |
duke@435 | 156 | // -6 [ saved r14 ] |
duke@435 | 157 | // -5 [ saved r13 ] |
duke@435 | 158 | // -4 [ saved r12 ] |
duke@435 | 159 | // -3 [ saved rdi ] |
duke@435 | 160 | // -2 [ saved rsi ] |
duke@435 | 161 | // -1 [ saved rbx ] |
duke@435 | 162 | // 0 [ saved rbp ] <--- rbp |
duke@435 | 163 | // 1 [ return address ] |
duke@435 | 164 | // 2 [ call wrapper ] |
duke@435 | 165 | // 3 [ result ] |
duke@435 | 166 | // 4 [ result type ] |
duke@435 | 167 | // 5 [ method ] |
duke@435 | 168 | // 6 [ entry point ] |
duke@435 | 169 | // 7 [ parameters ] |
duke@435 | 170 | // 8 [ parameter size ] |
duke@435 | 171 | // 9 [ thread ] |
duke@435 | 172 | // |
duke@435 | 173 | // Windows reserves the callers stack space for arguments 1-4. |
duke@435 | 174 | // We spill c_rarg0-c_rarg3 to this space. |
duke@435 | 175 | |
duke@435 | 176 | // Call stub stack layout word offsets from rbp |
duke@435 | 177 | enum call_stub_layout { |
duke@435 | 178 | #ifdef _WIN64 |
iveresov@2689 | 179 | xmm_save_first = 6, // save from xmm6 |
iveresov@2689 | 180 | xmm_save_last = 15, // to xmm15 |
iveresov@2689 | 181 | xmm_save_base = -9, |
iveresov@2689 | 182 | rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27 |
iveresov@2689 | 183 | r15_off = -7, |
duke@435 | 184 | r14_off = -6, |
duke@435 | 185 | r13_off = -5, |
duke@435 | 186 | r12_off = -4, |
duke@435 | 187 | rdi_off = -3, |
duke@435 | 188 | rsi_off = -2, |
duke@435 | 189 | rbx_off = -1, |
duke@435 | 190 | rbp_off = 0, |
duke@435 | 191 | retaddr_off = 1, |
duke@435 | 192 | call_wrapper_off = 2, |
duke@435 | 193 | result_off = 3, |
duke@435 | 194 | result_type_off = 4, |
duke@435 | 195 | method_off = 5, |
duke@435 | 196 | entry_point_off = 6, |
duke@435 | 197 | parameters_off = 7, |
duke@435 | 198 | parameter_size_off = 8, |
duke@435 | 199 | thread_off = 9 |
duke@435 | 200 | #else |
duke@435 | 201 | rsp_after_call_off = -12, |
duke@435 | 202 | mxcsr_off = rsp_after_call_off, |
duke@435 | 203 | r15_off = -11, |
duke@435 | 204 | r14_off = -10, |
duke@435 | 205 | r13_off = -9, |
duke@435 | 206 | r12_off = -8, |
duke@435 | 207 | rbx_off = -7, |
duke@435 | 208 | call_wrapper_off = -6, |
duke@435 | 209 | result_off = -5, |
duke@435 | 210 | result_type_off = -4, |
duke@435 | 211 | method_off = -3, |
duke@435 | 212 | entry_point_off = -2, |
duke@435 | 213 | parameters_off = -1, |
duke@435 | 214 | rbp_off = 0, |
duke@435 | 215 | retaddr_off = 1, |
duke@435 | 216 | parameter_size_off = 2, |
duke@435 | 217 | thread_off = 3 |
duke@435 | 218 | #endif |
duke@435 | 219 | }; |
duke@435 | 220 | |
iveresov@2689 | 221 | #ifdef _WIN64 |
iveresov@2689 | 222 | Address xmm_save(int reg) { |
iveresov@2689 | 223 | assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range"); |
iveresov@2689 | 224 | return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize); |
iveresov@2689 | 225 | } |
iveresov@2689 | 226 | #endif |
iveresov@2689 | 227 | |
duke@435 | 228 | address generate_call_stub(address& return_address) { |
duke@435 | 229 | assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 && |
duke@435 | 230 | (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off, |
duke@435 | 231 | "adjust this code"); |
duke@435 | 232 | StubCodeMark mark(this, "StubRoutines", "call_stub"); |
duke@435 | 233 | address start = __ pc(); |
duke@435 | 234 | |
duke@435 | 235 | // same as in generate_catch_exception()! |
duke@435 | 236 | const Address rsp_after_call(rbp, rsp_after_call_off * wordSize); |
duke@435 | 237 | |
duke@435 | 238 | const Address call_wrapper (rbp, call_wrapper_off * wordSize); |
duke@435 | 239 | const Address result (rbp, result_off * wordSize); |
duke@435 | 240 | const Address result_type (rbp, result_type_off * wordSize); |
duke@435 | 241 | const Address method (rbp, method_off * wordSize); |
duke@435 | 242 | const Address entry_point (rbp, entry_point_off * wordSize); |
duke@435 | 243 | const Address parameters (rbp, parameters_off * wordSize); |
duke@435 | 244 | const Address parameter_size(rbp, parameter_size_off * wordSize); |
duke@435 | 245 | |
duke@435 | 246 | // same as in generate_catch_exception()! |
duke@435 | 247 | const Address thread (rbp, thread_off * wordSize); |
duke@435 | 248 | |
duke@435 | 249 | const Address r15_save(rbp, r15_off * wordSize); |
duke@435 | 250 | const Address r14_save(rbp, r14_off * wordSize); |
duke@435 | 251 | const Address r13_save(rbp, r13_off * wordSize); |
duke@435 | 252 | const Address r12_save(rbp, r12_off * wordSize); |
duke@435 | 253 | const Address rbx_save(rbp, rbx_off * wordSize); |
duke@435 | 254 | |
duke@435 | 255 | // stub code |
duke@435 | 256 | __ enter(); |
never@739 | 257 | __ subptr(rsp, -rsp_after_call_off * wordSize); |
duke@435 | 258 | |
duke@435 | 259 | // save register parameters |
duke@435 | 260 | #ifndef _WIN64 |
never@739 | 261 | __ movptr(parameters, c_rarg5); // parameters |
never@739 | 262 | __ movptr(entry_point, c_rarg4); // entry_point |
duke@435 | 263 | #endif |
duke@435 | 264 | |
never@739 | 265 | __ movptr(method, c_rarg3); // method |
never@739 | 266 | __ movl(result_type, c_rarg2); // result type |
never@739 | 267 | __ movptr(result, c_rarg1); // result |
never@739 | 268 | __ movptr(call_wrapper, c_rarg0); // call wrapper |
duke@435 | 269 | |
duke@435 | 270 | // save regs belonging to calling function |
never@739 | 271 | __ movptr(rbx_save, rbx); |
never@739 | 272 | __ movptr(r12_save, r12); |
never@739 | 273 | __ movptr(r13_save, r13); |
never@739 | 274 | __ movptr(r14_save, r14); |
never@739 | 275 | __ movptr(r15_save, r15); |
duke@435 | 276 | #ifdef _WIN64 |
iveresov@2689 | 277 | for (int i = 6; i <= 15; i++) { |
iveresov@2689 | 278 | __ movdqu(xmm_save(i), as_XMMRegister(i)); |
iveresov@2689 | 279 | } |
iveresov@2689 | 280 | |
duke@435 | 281 | const Address rdi_save(rbp, rdi_off * wordSize); |
duke@435 | 282 | const Address rsi_save(rbp, rsi_off * wordSize); |
duke@435 | 283 | |
never@739 | 284 | __ movptr(rsi_save, rsi); |
never@739 | 285 | __ movptr(rdi_save, rdi); |
duke@435 | 286 | #else |
duke@435 | 287 | const Address mxcsr_save(rbp, mxcsr_off * wordSize); |
duke@435 | 288 | { |
duke@435 | 289 | Label skip_ldmx; |
duke@435 | 290 | __ stmxcsr(mxcsr_save); |
duke@435 | 291 | __ movl(rax, mxcsr_save); |
duke@435 | 292 | __ andl(rax, MXCSR_MASK); // Only check control and mask bits |
never@739 | 293 | ExternalAddress mxcsr_std(StubRoutines::x86::mxcsr_std()); |
duke@435 | 294 | __ cmp32(rax, mxcsr_std); |
duke@435 | 295 | __ jcc(Assembler::equal, skip_ldmx); |
duke@435 | 296 | __ ldmxcsr(mxcsr_std); |
duke@435 | 297 | __ bind(skip_ldmx); |
duke@435 | 298 | } |
duke@435 | 299 | #endif |
duke@435 | 300 | |
duke@435 | 301 | // Load up thread register |
never@739 | 302 | __ movptr(r15_thread, thread); |
coleenp@548 | 303 | __ reinit_heapbase(); |
duke@435 | 304 | |
duke@435 | 305 | #ifdef ASSERT |
duke@435 | 306 | // make sure we have no pending exceptions |
duke@435 | 307 | { |
duke@435 | 308 | Label L; |
never@739 | 309 | __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
duke@435 | 310 | __ jcc(Assembler::equal, L); |
duke@435 | 311 | __ stop("StubRoutines::call_stub: entered with pending exception"); |
duke@435 | 312 | __ bind(L); |
duke@435 | 313 | } |
duke@435 | 314 | #endif |
duke@435 | 315 | |
duke@435 | 316 | // pass parameters if any |
duke@435 | 317 | BLOCK_COMMENT("pass parameters if any"); |
duke@435 | 318 | Label parameters_done; |
duke@435 | 319 | __ movl(c_rarg3, parameter_size); |
duke@435 | 320 | __ testl(c_rarg3, c_rarg3); |
duke@435 | 321 | __ jcc(Assembler::zero, parameters_done); |
duke@435 | 322 | |
duke@435 | 323 | Label loop; |
never@739 | 324 | __ movptr(c_rarg2, parameters); // parameter pointer |
never@739 | 325 | __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1 |
duke@435 | 326 | __ BIND(loop); |
never@739 | 327 | __ movptr(rax, Address(c_rarg2, 0));// get parameter |
never@739 | 328 | __ addptr(c_rarg2, wordSize); // advance to next parameter |
never@739 | 329 | __ decrementl(c_rarg1); // decrement counter |
never@739 | 330 | __ push(rax); // pass parameter |
duke@435 | 331 | __ jcc(Assembler::notZero, loop); |
duke@435 | 332 | |
duke@435 | 333 | // call Java function |
duke@435 | 334 | __ BIND(parameters_done); |
never@739 | 335 | __ movptr(rbx, method); // get methodOop |
never@739 | 336 | __ movptr(c_rarg1, entry_point); // get entry_point |
never@739 | 337 | __ mov(r13, rsp); // set sender sp |
duke@435 | 338 | BLOCK_COMMENT("call Java function"); |
duke@435 | 339 | __ call(c_rarg1); |
duke@435 | 340 | |
duke@435 | 341 | BLOCK_COMMENT("call_stub_return_address:"); |
duke@435 | 342 | return_address = __ pc(); |
duke@435 | 343 | |
duke@435 | 344 | // store result depending on type (everything that is not |
duke@435 | 345 | // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT) |
never@739 | 346 | __ movptr(c_rarg0, result); |
duke@435 | 347 | Label is_long, is_float, is_double, exit; |
duke@435 | 348 | __ movl(c_rarg1, result_type); |
duke@435 | 349 | __ cmpl(c_rarg1, T_OBJECT); |
duke@435 | 350 | __ jcc(Assembler::equal, is_long); |
duke@435 | 351 | __ cmpl(c_rarg1, T_LONG); |
duke@435 | 352 | __ jcc(Assembler::equal, is_long); |
duke@435 | 353 | __ cmpl(c_rarg1, T_FLOAT); |
duke@435 | 354 | __ jcc(Assembler::equal, is_float); |
duke@435 | 355 | __ cmpl(c_rarg1, T_DOUBLE); |
duke@435 | 356 | __ jcc(Assembler::equal, is_double); |
duke@435 | 357 | |
duke@435 | 358 | // handle T_INT case |
duke@435 | 359 | __ movl(Address(c_rarg0, 0), rax); |
duke@435 | 360 | |
duke@435 | 361 | __ BIND(exit); |
duke@435 | 362 | |
duke@435 | 363 | // pop parameters |
never@739 | 364 | __ lea(rsp, rsp_after_call); |
duke@435 | 365 | |
duke@435 | 366 | #ifdef ASSERT |
duke@435 | 367 | // verify that threads correspond |
duke@435 | 368 | { |
duke@435 | 369 | Label L, S; |
never@739 | 370 | __ cmpptr(r15_thread, thread); |
duke@435 | 371 | __ jcc(Assembler::notEqual, S); |
duke@435 | 372 | __ get_thread(rbx); |
never@739 | 373 | __ cmpptr(r15_thread, rbx); |
duke@435 | 374 | __ jcc(Assembler::equal, L); |
duke@435 | 375 | __ bind(S); |
duke@435 | 376 | __ jcc(Assembler::equal, L); |
duke@435 | 377 | __ stop("StubRoutines::call_stub: threads must correspond"); |
duke@435 | 378 | __ bind(L); |
duke@435 | 379 | } |
duke@435 | 380 | #endif |
duke@435 | 381 | |
duke@435 | 382 | // restore regs belonging to calling function |
iveresov@2689 | 383 | #ifdef _WIN64 |
iveresov@2689 | 384 | for (int i = 15; i >= 6; i--) { |
iveresov@2689 | 385 | __ movdqu(as_XMMRegister(i), xmm_save(i)); |
iveresov@2689 | 386 | } |
iveresov@2689 | 387 | #endif |
never@739 | 388 | __ movptr(r15, r15_save); |
never@739 | 389 | __ movptr(r14, r14_save); |
never@739 | 390 | __ movptr(r13, r13_save); |
never@739 | 391 | __ movptr(r12, r12_save); |
never@739 | 392 | __ movptr(rbx, rbx_save); |
duke@435 | 393 | |
duke@435 | 394 | #ifdef _WIN64 |
never@739 | 395 | __ movptr(rdi, rdi_save); |
never@739 | 396 | __ movptr(rsi, rsi_save); |
duke@435 | 397 | #else |
duke@435 | 398 | __ ldmxcsr(mxcsr_save); |
duke@435 | 399 | #endif |
duke@435 | 400 | |
duke@435 | 401 | // restore rsp |
never@739 | 402 | __ addptr(rsp, -rsp_after_call_off * wordSize); |
duke@435 | 403 | |
duke@435 | 404 | // return |
never@739 | 405 | __ pop(rbp); |
duke@435 | 406 | __ ret(0); |
duke@435 | 407 | |
duke@435 | 408 | // handle return types different from T_INT |
duke@435 | 409 | __ BIND(is_long); |
duke@435 | 410 | __ movq(Address(c_rarg0, 0), rax); |
duke@435 | 411 | __ jmp(exit); |
duke@435 | 412 | |
duke@435 | 413 | __ BIND(is_float); |
duke@435 | 414 | __ movflt(Address(c_rarg0, 0), xmm0); |
duke@435 | 415 | __ jmp(exit); |
duke@435 | 416 | |
duke@435 | 417 | __ BIND(is_double); |
duke@435 | 418 | __ movdbl(Address(c_rarg0, 0), xmm0); |
duke@435 | 419 | __ jmp(exit); |
duke@435 | 420 | |
duke@435 | 421 | return start; |
duke@435 | 422 | } |
duke@435 | 423 | |
duke@435 | 424 | // Return point for a Java call if there's an exception thrown in |
duke@435 | 425 | // Java code. The exception is caught and transformed into a |
duke@435 | 426 | // pending exception stored in JavaThread that can be tested from |
duke@435 | 427 | // within the VM. |
duke@435 | 428 | // |
duke@435 | 429 | // Note: Usually the parameters are removed by the callee. In case |
duke@435 | 430 | // of an exception crossing an activation frame boundary, that is |
duke@435 | 431 | // not the case if the callee is compiled code => need to setup the |
duke@435 | 432 | // rsp. |
duke@435 | 433 | // |
duke@435 | 434 | // rax: exception oop |
duke@435 | 435 | |
duke@435 | 436 | address generate_catch_exception() { |
duke@435 | 437 | StubCodeMark mark(this, "StubRoutines", "catch_exception"); |
duke@435 | 438 | address start = __ pc(); |
duke@435 | 439 | |
duke@435 | 440 | // same as in generate_call_stub(): |
duke@435 | 441 | const Address rsp_after_call(rbp, rsp_after_call_off * wordSize); |
duke@435 | 442 | const Address thread (rbp, thread_off * wordSize); |
duke@435 | 443 | |
duke@435 | 444 | #ifdef ASSERT |
duke@435 | 445 | // verify that threads correspond |
duke@435 | 446 | { |
duke@435 | 447 | Label L, S; |
never@739 | 448 | __ cmpptr(r15_thread, thread); |
duke@435 | 449 | __ jcc(Assembler::notEqual, S); |
duke@435 | 450 | __ get_thread(rbx); |
never@739 | 451 | __ cmpptr(r15_thread, rbx); |
duke@435 | 452 | __ jcc(Assembler::equal, L); |
duke@435 | 453 | __ bind(S); |
duke@435 | 454 | __ stop("StubRoutines::catch_exception: threads must correspond"); |
duke@435 | 455 | __ bind(L); |
duke@435 | 456 | } |
duke@435 | 457 | #endif |
duke@435 | 458 | |
duke@435 | 459 | // set pending exception |
duke@435 | 460 | __ verify_oop(rax); |
duke@435 | 461 | |
never@739 | 462 | __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax); |
duke@435 | 463 | __ lea(rscratch1, ExternalAddress((address)__FILE__)); |
never@739 | 464 | __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1); |
duke@435 | 465 | __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__); |
duke@435 | 466 | |
duke@435 | 467 | // complete return to VM |
duke@435 | 468 | assert(StubRoutines::_call_stub_return_address != NULL, |
duke@435 | 469 | "_call_stub_return_address must have been generated before"); |
duke@435 | 470 | __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address)); |
duke@435 | 471 | |
duke@435 | 472 | return start; |
duke@435 | 473 | } |
duke@435 | 474 | |
duke@435 | 475 | // Continuation point for runtime calls returning with a pending |
duke@435 | 476 | // exception. The pending exception check happened in the runtime |
duke@435 | 477 | // or native call stub. The pending exception in Thread is |
duke@435 | 478 | // converted into a Java-level exception. |
duke@435 | 479 | // |
duke@435 | 480 | // Contract with Java-level exception handlers: |
duke@435 | 481 | // rax: exception |
duke@435 | 482 | // rdx: throwing pc |
duke@435 | 483 | // |
duke@435 | 484 | // NOTE: At entry of this stub, exception-pc must be on stack !! |
duke@435 | 485 | |
duke@435 | 486 | address generate_forward_exception() { |
duke@435 | 487 | StubCodeMark mark(this, "StubRoutines", "forward exception"); |
duke@435 | 488 | address start = __ pc(); |
duke@435 | 489 | |
duke@435 | 490 | // Upon entry, the sp points to the return address returning into |
duke@435 | 491 | // Java (interpreted or compiled) code; i.e., the return address |
duke@435 | 492 | // becomes the throwing pc. |
duke@435 | 493 | // |
duke@435 | 494 | // Arguments pushed before the runtime call are still on the stack |
duke@435 | 495 | // but the exception handler will reset the stack pointer -> |
duke@435 | 496 | // ignore them. A potential result in registers can be ignored as |
duke@435 | 497 | // well. |
duke@435 | 498 | |
duke@435 | 499 | #ifdef ASSERT |
duke@435 | 500 | // make sure this code is only executed if there is a pending exception |
duke@435 | 501 | { |
duke@435 | 502 | Label L; |
never@739 | 503 | __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL); |
duke@435 | 504 | __ jcc(Assembler::notEqual, L); |
duke@435 | 505 | __ stop("StubRoutines::forward exception: no pending exception (1)"); |
duke@435 | 506 | __ bind(L); |
duke@435 | 507 | } |
duke@435 | 508 | #endif |
duke@435 | 509 | |
duke@435 | 510 | // compute exception handler into rbx |
never@739 | 511 | __ movptr(c_rarg0, Address(rsp, 0)); |
duke@435 | 512 | BLOCK_COMMENT("call exception_handler_for_return_address"); |
duke@435 | 513 | __ call_VM_leaf(CAST_FROM_FN_PTR(address, |
duke@435 | 514 | SharedRuntime::exception_handler_for_return_address), |
twisti@1730 | 515 | r15_thread, c_rarg0); |
never@739 | 516 | __ mov(rbx, rax); |
duke@435 | 517 | |
duke@435 | 518 | // setup rax & rdx, remove return address & clear pending exception |
never@739 | 519 | __ pop(rdx); |
never@739 | 520 | __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset())); |
xlu@947 | 521 | __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
duke@435 | 522 | |
duke@435 | 523 | #ifdef ASSERT |
duke@435 | 524 | // make sure exception is set |
duke@435 | 525 | { |
duke@435 | 526 | Label L; |
never@739 | 527 | __ testptr(rax, rax); |
duke@435 | 528 | __ jcc(Assembler::notEqual, L); |
duke@435 | 529 | __ stop("StubRoutines::forward exception: no pending exception (2)"); |
duke@435 | 530 | __ bind(L); |
duke@435 | 531 | } |
duke@435 | 532 | #endif |
duke@435 | 533 | |
duke@435 | 534 | // continue at exception handler (return address removed) |
duke@435 | 535 | // rax: exception |
duke@435 | 536 | // rbx: exception handler |
duke@435 | 537 | // rdx: throwing pc |
duke@435 | 538 | __ verify_oop(rax); |
duke@435 | 539 | __ jmp(rbx); |
duke@435 | 540 | |
duke@435 | 541 | return start; |
duke@435 | 542 | } |
duke@435 | 543 | |
duke@435 | 544 | // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest) |
duke@435 | 545 | // |
duke@435 | 546 | // Arguments : |
duke@435 | 547 | // c_rarg0: exchange_value |
duke@435 | 548 | // c_rarg0: dest |
duke@435 | 549 | // |
duke@435 | 550 | // Result: |
duke@435 | 551 | // *dest <- ex, return (orig *dest) |
duke@435 | 552 | address generate_atomic_xchg() { |
duke@435 | 553 | StubCodeMark mark(this, "StubRoutines", "atomic_xchg"); |
duke@435 | 554 | address start = __ pc(); |
duke@435 | 555 | |
duke@435 | 556 | __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow |
duke@435 | 557 | __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK |
duke@435 | 558 | __ ret(0); |
duke@435 | 559 | |
duke@435 | 560 | return start; |
duke@435 | 561 | } |
duke@435 | 562 | |
duke@435 | 563 | // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest) |
duke@435 | 564 | // |
duke@435 | 565 | // Arguments : |
duke@435 | 566 | // c_rarg0: exchange_value |
duke@435 | 567 | // c_rarg1: dest |
duke@435 | 568 | // |
duke@435 | 569 | // Result: |
duke@435 | 570 | // *dest <- ex, return (orig *dest) |
duke@435 | 571 | address generate_atomic_xchg_ptr() { |
duke@435 | 572 | StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr"); |
duke@435 | 573 | address start = __ pc(); |
duke@435 | 574 | |
never@739 | 575 | __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow |
never@739 | 576 | __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK |
duke@435 | 577 | __ ret(0); |
duke@435 | 578 | |
duke@435 | 579 | return start; |
duke@435 | 580 | } |
duke@435 | 581 | |
duke@435 | 582 | // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest, |
duke@435 | 583 | // jint compare_value) |
duke@435 | 584 | // |
duke@435 | 585 | // Arguments : |
duke@435 | 586 | // c_rarg0: exchange_value |
duke@435 | 587 | // c_rarg1: dest |
duke@435 | 588 | // c_rarg2: compare_value |
duke@435 | 589 | // |
duke@435 | 590 | // Result: |
duke@435 | 591 | // if ( compare_value == *dest ) { |
duke@435 | 592 | // *dest = exchange_value |
duke@435 | 593 | // return compare_value; |
duke@435 | 594 | // else |
duke@435 | 595 | // return *dest; |
duke@435 | 596 | address generate_atomic_cmpxchg() { |
duke@435 | 597 | StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg"); |
duke@435 | 598 | address start = __ pc(); |
duke@435 | 599 | |
duke@435 | 600 | __ movl(rax, c_rarg2); |
duke@435 | 601 | if ( os::is_MP() ) __ lock(); |
duke@435 | 602 | __ cmpxchgl(c_rarg0, Address(c_rarg1, 0)); |
duke@435 | 603 | __ ret(0); |
duke@435 | 604 | |
duke@435 | 605 | return start; |
duke@435 | 606 | } |
duke@435 | 607 | |
duke@435 | 608 | // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value, |
duke@435 | 609 | // volatile jlong* dest, |
duke@435 | 610 | // jlong compare_value) |
duke@435 | 611 | // Arguments : |
duke@435 | 612 | // c_rarg0: exchange_value |
duke@435 | 613 | // c_rarg1: dest |
duke@435 | 614 | // c_rarg2: compare_value |
duke@435 | 615 | // |
duke@435 | 616 | // Result: |
duke@435 | 617 | // if ( compare_value == *dest ) { |
duke@435 | 618 | // *dest = exchange_value |
duke@435 | 619 | // return compare_value; |
duke@435 | 620 | // else |
duke@435 | 621 | // return *dest; |
duke@435 | 622 | address generate_atomic_cmpxchg_long() { |
duke@435 | 623 | StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long"); |
duke@435 | 624 | address start = __ pc(); |
duke@435 | 625 | |
duke@435 | 626 | __ movq(rax, c_rarg2); |
duke@435 | 627 | if ( os::is_MP() ) __ lock(); |
duke@435 | 628 | __ cmpxchgq(c_rarg0, Address(c_rarg1, 0)); |
duke@435 | 629 | __ ret(0); |
duke@435 | 630 | |
duke@435 | 631 | return start; |
duke@435 | 632 | } |
duke@435 | 633 | |
duke@435 | 634 | // Support for jint atomic::add(jint add_value, volatile jint* dest) |
duke@435 | 635 | // |
duke@435 | 636 | // Arguments : |
duke@435 | 637 | // c_rarg0: add_value |
duke@435 | 638 | // c_rarg1: dest |
duke@435 | 639 | // |
duke@435 | 640 | // Result: |
duke@435 | 641 | // *dest += add_value |
duke@435 | 642 | // return *dest; |
duke@435 | 643 | address generate_atomic_add() { |
duke@435 | 644 | StubCodeMark mark(this, "StubRoutines", "atomic_add"); |
duke@435 | 645 | address start = __ pc(); |
duke@435 | 646 | |
duke@435 | 647 | __ movl(rax, c_rarg0); |
duke@435 | 648 | if ( os::is_MP() ) __ lock(); |
duke@435 | 649 | __ xaddl(Address(c_rarg1, 0), c_rarg0); |
duke@435 | 650 | __ addl(rax, c_rarg0); |
duke@435 | 651 | __ ret(0); |
duke@435 | 652 | |
duke@435 | 653 | return start; |
duke@435 | 654 | } |
duke@435 | 655 | |
duke@435 | 656 | // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest) |
duke@435 | 657 | // |
duke@435 | 658 | // Arguments : |
duke@435 | 659 | // c_rarg0: add_value |
duke@435 | 660 | // c_rarg1: dest |
duke@435 | 661 | // |
duke@435 | 662 | // Result: |
duke@435 | 663 | // *dest += add_value |
duke@435 | 664 | // return *dest; |
duke@435 | 665 | address generate_atomic_add_ptr() { |
duke@435 | 666 | StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr"); |
duke@435 | 667 | address start = __ pc(); |
duke@435 | 668 | |
never@739 | 669 | __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow |
duke@435 | 670 | if ( os::is_MP() ) __ lock(); |
never@739 | 671 | __ xaddptr(Address(c_rarg1, 0), c_rarg0); |
never@739 | 672 | __ addptr(rax, c_rarg0); |
duke@435 | 673 | __ ret(0); |
duke@435 | 674 | |
duke@435 | 675 | return start; |
duke@435 | 676 | } |
duke@435 | 677 | |
duke@435 | 678 | // Support for intptr_t OrderAccess::fence() |
duke@435 | 679 | // |
duke@435 | 680 | // Arguments : |
duke@435 | 681 | // |
duke@435 | 682 | // Result: |
duke@435 | 683 | address generate_orderaccess_fence() { |
duke@435 | 684 | StubCodeMark mark(this, "StubRoutines", "orderaccess_fence"); |
duke@435 | 685 | address start = __ pc(); |
never@1106 | 686 | __ membar(Assembler::StoreLoad); |
duke@435 | 687 | __ ret(0); |
duke@435 | 688 | |
duke@435 | 689 | return start; |
duke@435 | 690 | } |
duke@435 | 691 | |
duke@435 | 692 | // Support for intptr_t get_previous_fp() |
duke@435 | 693 | // |
duke@435 | 694 | // This routine is used to find the previous frame pointer for the |
duke@435 | 695 | // caller (current_frame_guess). This is used as part of debugging |
duke@435 | 696 | // ps() is seemingly lost trying to find frames. |
duke@435 | 697 | // This code assumes that caller current_frame_guess) has a frame. |
duke@435 | 698 | address generate_get_previous_fp() { |
duke@435 | 699 | StubCodeMark mark(this, "StubRoutines", "get_previous_fp"); |
duke@435 | 700 | const Address old_fp(rbp, 0); |
duke@435 | 701 | const Address older_fp(rax, 0); |
duke@435 | 702 | address start = __ pc(); |
duke@435 | 703 | |
duke@435 | 704 | __ enter(); |
never@739 | 705 | __ movptr(rax, old_fp); // callers fp |
never@739 | 706 | __ movptr(rax, older_fp); // the frame for ps() |
never@739 | 707 | __ pop(rbp); |
duke@435 | 708 | __ ret(0); |
duke@435 | 709 | |
duke@435 | 710 | return start; |
duke@435 | 711 | } |
duke@435 | 712 | |
duke@435 | 713 | //---------------------------------------------------------------------------------------------------- |
duke@435 | 714 | // Support for void verify_mxcsr() |
duke@435 | 715 | // |
duke@435 | 716 | // This routine is used with -Xcheck:jni to verify that native |
duke@435 | 717 | // JNI code does not return to Java code without restoring the |
duke@435 | 718 | // MXCSR register to our expected state. |
duke@435 | 719 | |
duke@435 | 720 | address generate_verify_mxcsr() { |
duke@435 | 721 | StubCodeMark mark(this, "StubRoutines", "verify_mxcsr"); |
duke@435 | 722 | address start = __ pc(); |
duke@435 | 723 | |
duke@435 | 724 | const Address mxcsr_save(rsp, 0); |
duke@435 | 725 | |
duke@435 | 726 | if (CheckJNICalls) { |
duke@435 | 727 | Label ok_ret; |
never@739 | 728 | __ push(rax); |
never@739 | 729 | __ subptr(rsp, wordSize); // allocate a temp location |
duke@435 | 730 | __ stmxcsr(mxcsr_save); |
duke@435 | 731 | __ movl(rax, mxcsr_save); |
duke@435 | 732 | __ andl(rax, MXCSR_MASK); // Only check control and mask bits |
never@739 | 733 | __ cmpl(rax, *(int *)(StubRoutines::x86::mxcsr_std())); |
duke@435 | 734 | __ jcc(Assembler::equal, ok_ret); |
duke@435 | 735 | |
duke@435 | 736 | __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall"); |
duke@435 | 737 | |
never@739 | 738 | __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std())); |
duke@435 | 739 | |
duke@435 | 740 | __ bind(ok_ret); |
never@739 | 741 | __ addptr(rsp, wordSize); |
never@739 | 742 | __ pop(rax); |
duke@435 | 743 | } |
duke@435 | 744 | |
duke@435 | 745 | __ ret(0); |
duke@435 | 746 | |
duke@435 | 747 | return start; |
duke@435 | 748 | } |
duke@435 | 749 | |
duke@435 | 750 | address generate_f2i_fixup() { |
duke@435 | 751 | StubCodeMark mark(this, "StubRoutines", "f2i_fixup"); |
duke@435 | 752 | Address inout(rsp, 5 * wordSize); // return address + 4 saves |
duke@435 | 753 | |
duke@435 | 754 | address start = __ pc(); |
duke@435 | 755 | |
duke@435 | 756 | Label L; |
duke@435 | 757 | |
never@739 | 758 | __ push(rax); |
never@739 | 759 | __ push(c_rarg3); |
never@739 | 760 | __ push(c_rarg2); |
never@739 | 761 | __ push(c_rarg1); |
duke@435 | 762 | |
duke@435 | 763 | __ movl(rax, 0x7f800000); |
duke@435 | 764 | __ xorl(c_rarg3, c_rarg3); |
duke@435 | 765 | __ movl(c_rarg2, inout); |
duke@435 | 766 | __ movl(c_rarg1, c_rarg2); |
duke@435 | 767 | __ andl(c_rarg1, 0x7fffffff); |
duke@435 | 768 | __ cmpl(rax, c_rarg1); // NaN? -> 0 |
duke@435 | 769 | __ jcc(Assembler::negative, L); |
duke@435 | 770 | __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint |
duke@435 | 771 | __ movl(c_rarg3, 0x80000000); |
duke@435 | 772 | __ movl(rax, 0x7fffffff); |
duke@435 | 773 | __ cmovl(Assembler::positive, c_rarg3, rax); |
duke@435 | 774 | |
duke@435 | 775 | __ bind(L); |
never@739 | 776 | __ movptr(inout, c_rarg3); |
never@739 | 777 | |
never@739 | 778 | __ pop(c_rarg1); |
never@739 | 779 | __ pop(c_rarg2); |
never@739 | 780 | __ pop(c_rarg3); |
never@739 | 781 | __ pop(rax); |
duke@435 | 782 | |
duke@435 | 783 | __ ret(0); |
duke@435 | 784 | |
duke@435 | 785 | return start; |
duke@435 | 786 | } |
duke@435 | 787 | |
duke@435 | 788 | address generate_f2l_fixup() { |
duke@435 | 789 | StubCodeMark mark(this, "StubRoutines", "f2l_fixup"); |
duke@435 | 790 | Address inout(rsp, 5 * wordSize); // return address + 4 saves |
duke@435 | 791 | address start = __ pc(); |
duke@435 | 792 | |
duke@435 | 793 | Label L; |
duke@435 | 794 | |
never@739 | 795 | __ push(rax); |
never@739 | 796 | __ push(c_rarg3); |
never@739 | 797 | __ push(c_rarg2); |
never@739 | 798 | __ push(c_rarg1); |
duke@435 | 799 | |
duke@435 | 800 | __ movl(rax, 0x7f800000); |
duke@435 | 801 | __ xorl(c_rarg3, c_rarg3); |
duke@435 | 802 | __ movl(c_rarg2, inout); |
duke@435 | 803 | __ movl(c_rarg1, c_rarg2); |
duke@435 | 804 | __ andl(c_rarg1, 0x7fffffff); |
duke@435 | 805 | __ cmpl(rax, c_rarg1); // NaN? -> 0 |
duke@435 | 806 | __ jcc(Assembler::negative, L); |
duke@435 | 807 | __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong |
duke@435 | 808 | __ mov64(c_rarg3, 0x8000000000000000); |
duke@435 | 809 | __ mov64(rax, 0x7fffffffffffffff); |
never@739 | 810 | __ cmov(Assembler::positive, c_rarg3, rax); |
duke@435 | 811 | |
duke@435 | 812 | __ bind(L); |
never@739 | 813 | __ movptr(inout, c_rarg3); |
never@739 | 814 | |
never@739 | 815 | __ pop(c_rarg1); |
never@739 | 816 | __ pop(c_rarg2); |
never@739 | 817 | __ pop(c_rarg3); |
never@739 | 818 | __ pop(rax); |
duke@435 | 819 | |
duke@435 | 820 | __ ret(0); |
duke@435 | 821 | |
duke@435 | 822 | return start; |
duke@435 | 823 | } |
duke@435 | 824 | |
duke@435 | 825 | address generate_d2i_fixup() { |
duke@435 | 826 | StubCodeMark mark(this, "StubRoutines", "d2i_fixup"); |
duke@435 | 827 | Address inout(rsp, 6 * wordSize); // return address + 5 saves |
duke@435 | 828 | |
duke@435 | 829 | address start = __ pc(); |
duke@435 | 830 | |
duke@435 | 831 | Label L; |
duke@435 | 832 | |
never@739 | 833 | __ push(rax); |
never@739 | 834 | __ push(c_rarg3); |
never@739 | 835 | __ push(c_rarg2); |
never@739 | 836 | __ push(c_rarg1); |
never@739 | 837 | __ push(c_rarg0); |
duke@435 | 838 | |
duke@435 | 839 | __ movl(rax, 0x7ff00000); |
duke@435 | 840 | __ movq(c_rarg2, inout); |
duke@435 | 841 | __ movl(c_rarg3, c_rarg2); |
never@739 | 842 | __ mov(c_rarg1, c_rarg2); |
never@739 | 843 | __ mov(c_rarg0, c_rarg2); |
duke@435 | 844 | __ negl(c_rarg3); |
never@739 | 845 | __ shrptr(c_rarg1, 0x20); |
duke@435 | 846 | __ orl(c_rarg3, c_rarg2); |
duke@435 | 847 | __ andl(c_rarg1, 0x7fffffff); |
duke@435 | 848 | __ xorl(c_rarg2, c_rarg2); |
duke@435 | 849 | __ shrl(c_rarg3, 0x1f); |
duke@435 | 850 | __ orl(c_rarg1, c_rarg3); |
duke@435 | 851 | __ cmpl(rax, c_rarg1); |
duke@435 | 852 | __ jcc(Assembler::negative, L); // NaN -> 0 |
never@739 | 853 | __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint |
duke@435 | 854 | __ movl(c_rarg2, 0x80000000); |
duke@435 | 855 | __ movl(rax, 0x7fffffff); |
never@739 | 856 | __ cmov(Assembler::positive, c_rarg2, rax); |
duke@435 | 857 | |
duke@435 | 858 | __ bind(L); |
never@739 | 859 | __ movptr(inout, c_rarg2); |
never@739 | 860 | |
never@739 | 861 | __ pop(c_rarg0); |
never@739 | 862 | __ pop(c_rarg1); |
never@739 | 863 | __ pop(c_rarg2); |
never@739 | 864 | __ pop(c_rarg3); |
never@739 | 865 | __ pop(rax); |
duke@435 | 866 | |
duke@435 | 867 | __ ret(0); |
duke@435 | 868 | |
duke@435 | 869 | return start; |
duke@435 | 870 | } |
duke@435 | 871 | |
duke@435 | 872 | address generate_d2l_fixup() { |
duke@435 | 873 | StubCodeMark mark(this, "StubRoutines", "d2l_fixup"); |
duke@435 | 874 | Address inout(rsp, 6 * wordSize); // return address + 5 saves |
duke@435 | 875 | |
duke@435 | 876 | address start = __ pc(); |
duke@435 | 877 | |
duke@435 | 878 | Label L; |
duke@435 | 879 | |
never@739 | 880 | __ push(rax); |
never@739 | 881 | __ push(c_rarg3); |
never@739 | 882 | __ push(c_rarg2); |
never@739 | 883 | __ push(c_rarg1); |
never@739 | 884 | __ push(c_rarg0); |
duke@435 | 885 | |
duke@435 | 886 | __ movl(rax, 0x7ff00000); |
duke@435 | 887 | __ movq(c_rarg2, inout); |
duke@435 | 888 | __ movl(c_rarg3, c_rarg2); |
never@739 | 889 | __ mov(c_rarg1, c_rarg2); |
never@739 | 890 | __ mov(c_rarg0, c_rarg2); |
duke@435 | 891 | __ negl(c_rarg3); |
never@739 | 892 | __ shrptr(c_rarg1, 0x20); |
duke@435 | 893 | __ orl(c_rarg3, c_rarg2); |
duke@435 | 894 | __ andl(c_rarg1, 0x7fffffff); |
duke@435 | 895 | __ xorl(c_rarg2, c_rarg2); |
duke@435 | 896 | __ shrl(c_rarg3, 0x1f); |
duke@435 | 897 | __ orl(c_rarg1, c_rarg3); |
duke@435 | 898 | __ cmpl(rax, c_rarg1); |
duke@435 | 899 | __ jcc(Assembler::negative, L); // NaN -> 0 |
duke@435 | 900 | __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong |
duke@435 | 901 | __ mov64(c_rarg2, 0x8000000000000000); |
duke@435 | 902 | __ mov64(rax, 0x7fffffffffffffff); |
duke@435 | 903 | __ cmovq(Assembler::positive, c_rarg2, rax); |
duke@435 | 904 | |
duke@435 | 905 | __ bind(L); |
duke@435 | 906 | __ movq(inout, c_rarg2); |
duke@435 | 907 | |
never@739 | 908 | __ pop(c_rarg0); |
never@739 | 909 | __ pop(c_rarg1); |
never@739 | 910 | __ pop(c_rarg2); |
never@739 | 911 | __ pop(c_rarg3); |
never@739 | 912 | __ pop(rax); |
duke@435 | 913 | |
duke@435 | 914 | __ ret(0); |
duke@435 | 915 | |
duke@435 | 916 | return start; |
duke@435 | 917 | } |
duke@435 | 918 | |
duke@435 | 919 | address generate_fp_mask(const char *stub_name, int64_t mask) { |
kvn@1800 | 920 | __ align(CodeEntryAlignment); |
duke@435 | 921 | StubCodeMark mark(this, "StubRoutines", stub_name); |
duke@435 | 922 | address start = __ pc(); |
duke@435 | 923 | |
duke@435 | 924 | __ emit_data64( mask, relocInfo::none ); |
duke@435 | 925 | __ emit_data64( mask, relocInfo::none ); |
duke@435 | 926 | |
duke@435 | 927 | return start; |
duke@435 | 928 | } |
duke@435 | 929 | |
duke@435 | 930 | // The following routine generates a subroutine to throw an |
duke@435 | 931 | // asynchronous UnknownError when an unsafe access gets a fault that |
duke@435 | 932 | // could not be reasonably prevented by the programmer. (Example: |
duke@435 | 933 | // SIGBUS/OBJERR.) |
duke@435 | 934 | address generate_handler_for_unsafe_access() { |
duke@435 | 935 | StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access"); |
duke@435 | 936 | address start = __ pc(); |
duke@435 | 937 | |
never@739 | 938 | __ push(0); // hole for return address-to-be |
never@739 | 939 | __ pusha(); // push registers |
duke@435 | 940 | Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord); |
duke@435 | 941 | |
never@3136 | 942 | // FIXME: this probably needs alignment logic |
never@3136 | 943 | |
never@739 | 944 | __ subptr(rsp, frame::arg_reg_save_area_bytes); |
duke@435 | 945 | BLOCK_COMMENT("call handle_unsafe_access"); |
duke@435 | 946 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access))); |
never@739 | 947 | __ addptr(rsp, frame::arg_reg_save_area_bytes); |
never@739 | 948 | |
never@739 | 949 | __ movptr(next_pc, rax); // stuff next address |
never@739 | 950 | __ popa(); |
duke@435 | 951 | __ ret(0); // jump to next address |
duke@435 | 952 | |
duke@435 | 953 | return start; |
duke@435 | 954 | } |
duke@435 | 955 | |
duke@435 | 956 | // Non-destructive plausibility checks for oops |
duke@435 | 957 | // |
duke@435 | 958 | // Arguments: |
duke@435 | 959 | // all args on stack! |
duke@435 | 960 | // |
duke@435 | 961 | // Stack after saving c_rarg3: |
duke@435 | 962 | // [tos + 0]: saved c_rarg3 |
duke@435 | 963 | // [tos + 1]: saved c_rarg2 |
kvn@559 | 964 | // [tos + 2]: saved r12 (several TemplateTable methods use it) |
kvn@559 | 965 | // [tos + 3]: saved flags |
kvn@559 | 966 | // [tos + 4]: return address |
kvn@559 | 967 | // * [tos + 5]: error message (char*) |
kvn@559 | 968 | // * [tos + 6]: object to verify (oop) |
kvn@559 | 969 | // * [tos + 7]: saved rax - saved by caller and bashed |
kvn@1938 | 970 | // * [tos + 8]: saved r10 (rscratch1) - saved by caller |
duke@435 | 971 | // * = popped on exit |
duke@435 | 972 | address generate_verify_oop() { |
duke@435 | 973 | StubCodeMark mark(this, "StubRoutines", "verify_oop"); |
duke@435 | 974 | address start = __ pc(); |
duke@435 | 975 | |
duke@435 | 976 | Label exit, error; |
duke@435 | 977 | |
never@739 | 978 | __ pushf(); |
duke@435 | 979 | __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr())); |
duke@435 | 980 | |
never@739 | 981 | __ push(r12); |
kvn@559 | 982 | |
duke@435 | 983 | // save c_rarg2 and c_rarg3 |
never@739 | 984 | __ push(c_rarg2); |
never@739 | 985 | __ push(c_rarg3); |
duke@435 | 986 | |
kvn@559 | 987 | enum { |
kvn@559 | 988 | // After previous pushes. |
kvn@559 | 989 | oop_to_verify = 6 * wordSize, |
kvn@559 | 990 | saved_rax = 7 * wordSize, |
kvn@1938 | 991 | saved_r10 = 8 * wordSize, |
kvn@559 | 992 | |
kvn@559 | 993 | // Before the call to MacroAssembler::debug(), see below. |
kvn@559 | 994 | return_addr = 16 * wordSize, |
kvn@559 | 995 | error_msg = 17 * wordSize |
kvn@559 | 996 | }; |
kvn@559 | 997 | |
duke@435 | 998 | // get object |
never@739 | 999 | __ movptr(rax, Address(rsp, oop_to_verify)); |
duke@435 | 1000 | |
duke@435 | 1001 | // make sure object is 'reasonable' |
never@739 | 1002 | __ testptr(rax, rax); |
duke@435 | 1003 | __ jcc(Assembler::zero, exit); // if obj is NULL it is OK |
duke@435 | 1004 | // Check if the oop is in the right area of memory |
never@739 | 1005 | __ movptr(c_rarg2, rax); |
xlu@947 | 1006 | __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask()); |
never@739 | 1007 | __ andptr(c_rarg2, c_rarg3); |
xlu@947 | 1008 | __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits()); |
never@739 | 1009 | __ cmpptr(c_rarg2, c_rarg3); |
duke@435 | 1010 | __ jcc(Assembler::notZero, error); |
duke@435 | 1011 | |
kvn@559 | 1012 | // set r12 to heapbase for load_klass() |
kvn@559 | 1013 | __ reinit_heapbase(); |
kvn@559 | 1014 | |
duke@435 | 1015 | // make sure klass is 'reasonable' |
coleenp@548 | 1016 | __ load_klass(rax, rax); // get klass |
never@739 | 1017 | __ testptr(rax, rax); |
duke@435 | 1018 | __ jcc(Assembler::zero, error); // if klass is NULL it is broken |
duke@435 | 1019 | // Check if the klass is in the right area of memory |
never@739 | 1020 | __ mov(c_rarg2, rax); |
xlu@947 | 1021 | __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask()); |
never@739 | 1022 | __ andptr(c_rarg2, c_rarg3); |
xlu@947 | 1023 | __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits()); |
never@739 | 1024 | __ cmpptr(c_rarg2, c_rarg3); |
duke@435 | 1025 | __ jcc(Assembler::notZero, error); |
duke@435 | 1026 | |
duke@435 | 1027 | // make sure klass' klass is 'reasonable' |
coleenp@548 | 1028 | __ load_klass(rax, rax); |
never@739 | 1029 | __ testptr(rax, rax); |
duke@435 | 1030 | __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken |
duke@435 | 1031 | // Check if the klass' klass is in the right area of memory |
xlu@947 | 1032 | __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask()); |
never@739 | 1033 | __ andptr(rax, c_rarg3); |
xlu@947 | 1034 | __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits()); |
never@739 | 1035 | __ cmpptr(rax, c_rarg3); |
duke@435 | 1036 | __ jcc(Assembler::notZero, error); |
duke@435 | 1037 | |
duke@435 | 1038 | // return if everything seems ok |
duke@435 | 1039 | __ bind(exit); |
never@739 | 1040 | __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back |
kvn@1938 | 1041 | __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back |
never@739 | 1042 | __ pop(c_rarg3); // restore c_rarg3 |
never@739 | 1043 | __ pop(c_rarg2); // restore c_rarg2 |
never@739 | 1044 | __ pop(r12); // restore r12 |
never@739 | 1045 | __ popf(); // restore flags |
kvn@1938 | 1046 | __ ret(4 * wordSize); // pop caller saved stuff |
duke@435 | 1047 | |
duke@435 | 1048 | // handle errors |
duke@435 | 1049 | __ bind(error); |
never@739 | 1050 | __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back |
kvn@1938 | 1051 | __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back |
never@739 | 1052 | __ pop(c_rarg3); // get saved c_rarg3 back |
never@739 | 1053 | __ pop(c_rarg2); // get saved c_rarg2 back |
never@739 | 1054 | __ pop(r12); // get saved r12 back |
never@739 | 1055 | __ popf(); // get saved flags off stack -- |
duke@435 | 1056 | // will be ignored |
duke@435 | 1057 | |
never@739 | 1058 | __ pusha(); // push registers |
duke@435 | 1059 | // (rip is already |
duke@435 | 1060 | // already pushed) |
kvn@559 | 1061 | // debug(char* msg, int64_t pc, int64_t regs[]) |
duke@435 | 1062 | // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and |
duke@435 | 1063 | // pushed all the registers, so now the stack looks like: |
duke@435 | 1064 | // [tos + 0] 16 saved registers |
duke@435 | 1065 | // [tos + 16] return address |
kvn@559 | 1066 | // * [tos + 17] error message (char*) |
kvn@559 | 1067 | // * [tos + 18] object to verify (oop) |
kvn@559 | 1068 | // * [tos + 19] saved rax - saved by caller and bashed |
kvn@1938 | 1069 | // * [tos + 20] saved r10 (rscratch1) - saved by caller |
kvn@559 | 1070 | // * = popped on exit |
kvn@559 | 1071 | |
never@739 | 1072 | __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message |
never@739 | 1073 | __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address |
never@739 | 1074 | __ movq(c_rarg2, rsp); // pass address of regs on stack |
never@739 | 1075 | __ mov(r12, rsp); // remember rsp |
never@739 | 1076 | __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows |
never@739 | 1077 | __ andptr(rsp, -16); // align stack as required by ABI |
duke@435 | 1078 | BLOCK_COMMENT("call MacroAssembler::debug"); |
never@739 | 1079 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64))); |
never@739 | 1080 | __ mov(rsp, r12); // restore rsp |
never@739 | 1081 | __ popa(); // pop registers (includes r12) |
kvn@1938 | 1082 | __ ret(4 * wordSize); // pop caller saved stuff |
duke@435 | 1083 | |
duke@435 | 1084 | return start; |
duke@435 | 1085 | } |
duke@435 | 1086 | |
duke@435 | 1087 | // |
duke@435 | 1088 | // Verify that a register contains clean 32-bits positive value |
duke@435 | 1089 | // (high 32-bits are 0) so it could be used in 64-bits shifts. |
duke@435 | 1090 | // |
duke@435 | 1091 | // Input: |
duke@435 | 1092 | // Rint - 32-bits value |
duke@435 | 1093 | // Rtmp - scratch |
duke@435 | 1094 | // |
duke@435 | 1095 | void assert_clean_int(Register Rint, Register Rtmp) { |
duke@435 | 1096 | #ifdef ASSERT |
duke@435 | 1097 | Label L; |
duke@435 | 1098 | assert_different_registers(Rtmp, Rint); |
duke@435 | 1099 | __ movslq(Rtmp, Rint); |
duke@435 | 1100 | __ cmpq(Rtmp, Rint); |
kvn@559 | 1101 | __ jcc(Assembler::equal, L); |
duke@435 | 1102 | __ stop("high 32-bits of int value are not 0"); |
duke@435 | 1103 | __ bind(L); |
duke@435 | 1104 | #endif |
duke@435 | 1105 | } |
duke@435 | 1106 | |
duke@435 | 1107 | // Generate overlap test for array copy stubs |
duke@435 | 1108 | // |
duke@435 | 1109 | // Input: |
duke@435 | 1110 | // c_rarg0 - from |
duke@435 | 1111 | // c_rarg1 - to |
duke@435 | 1112 | // c_rarg2 - element count |
duke@435 | 1113 | // |
duke@435 | 1114 | // Output: |
duke@435 | 1115 | // rax - &from[element count - 1] |
duke@435 | 1116 | // |
duke@435 | 1117 | void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) { |
duke@435 | 1118 | assert(no_overlap_target != NULL, "must be generated"); |
duke@435 | 1119 | array_overlap_test(no_overlap_target, NULL, sf); |
duke@435 | 1120 | } |
duke@435 | 1121 | void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) { |
duke@435 | 1122 | array_overlap_test(NULL, &L_no_overlap, sf); |
duke@435 | 1123 | } |
duke@435 | 1124 | void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) { |
duke@435 | 1125 | const Register from = c_rarg0; |
duke@435 | 1126 | const Register to = c_rarg1; |
duke@435 | 1127 | const Register count = c_rarg2; |
duke@435 | 1128 | const Register end_from = rax; |
duke@435 | 1129 | |
never@739 | 1130 | __ cmpptr(to, from); |
never@739 | 1131 | __ lea(end_from, Address(from, count, sf, 0)); |
duke@435 | 1132 | if (NOLp == NULL) { |
duke@435 | 1133 | ExternalAddress no_overlap(no_overlap_target); |
duke@435 | 1134 | __ jump_cc(Assembler::belowEqual, no_overlap); |
never@739 | 1135 | __ cmpptr(to, end_from); |
duke@435 | 1136 | __ jump_cc(Assembler::aboveEqual, no_overlap); |
duke@435 | 1137 | } else { |
duke@435 | 1138 | __ jcc(Assembler::belowEqual, (*NOLp)); |
never@739 | 1139 | __ cmpptr(to, end_from); |
duke@435 | 1140 | __ jcc(Assembler::aboveEqual, (*NOLp)); |
duke@435 | 1141 | } |
duke@435 | 1142 | } |
duke@435 | 1143 | |
duke@435 | 1144 | // Shuffle first three arg regs on Windows into Linux/Solaris locations. |
duke@435 | 1145 | // |
duke@435 | 1146 | // Outputs: |
duke@435 | 1147 | // rdi - rcx |
duke@435 | 1148 | // rsi - rdx |
duke@435 | 1149 | // rdx - r8 |
duke@435 | 1150 | // rcx - r9 |
duke@435 | 1151 | // |
duke@435 | 1152 | // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter |
duke@435 | 1153 | // are non-volatile. r9 and r10 should not be used by the caller. |
duke@435 | 1154 | // |
duke@435 | 1155 | void setup_arg_regs(int nargs = 3) { |
duke@435 | 1156 | const Register saved_rdi = r9; |
duke@435 | 1157 | const Register saved_rsi = r10; |
duke@435 | 1158 | assert(nargs == 3 || nargs == 4, "else fix"); |
duke@435 | 1159 | #ifdef _WIN64 |
duke@435 | 1160 | assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9, |
duke@435 | 1161 | "unexpected argument registers"); |
duke@435 | 1162 | if (nargs >= 4) |
never@739 | 1163 | __ mov(rax, r9); // r9 is also saved_rdi |
never@739 | 1164 | __ movptr(saved_rdi, rdi); |
never@739 | 1165 | __ movptr(saved_rsi, rsi); |
never@739 | 1166 | __ mov(rdi, rcx); // c_rarg0 |
never@739 | 1167 | __ mov(rsi, rdx); // c_rarg1 |
never@739 | 1168 | __ mov(rdx, r8); // c_rarg2 |
duke@435 | 1169 | if (nargs >= 4) |
never@739 | 1170 | __ mov(rcx, rax); // c_rarg3 (via rax) |
duke@435 | 1171 | #else |
duke@435 | 1172 | assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx, |
duke@435 | 1173 | "unexpected argument registers"); |
duke@435 | 1174 | #endif |
duke@435 | 1175 | } |
duke@435 | 1176 | |
duke@435 | 1177 | void restore_arg_regs() { |
duke@435 | 1178 | const Register saved_rdi = r9; |
duke@435 | 1179 | const Register saved_rsi = r10; |
duke@435 | 1180 | #ifdef _WIN64 |
never@739 | 1181 | __ movptr(rdi, saved_rdi); |
never@739 | 1182 | __ movptr(rsi, saved_rsi); |
duke@435 | 1183 | #endif |
duke@435 | 1184 | } |
duke@435 | 1185 | |
duke@435 | 1186 | // Generate code for an array write pre barrier |
duke@435 | 1187 | // |
duke@435 | 1188 | // addr - starting address |
iveresov@2606 | 1189 | // count - element count |
iveresov@2606 | 1190 | // tmp - scratch register |
duke@435 | 1191 | // |
duke@435 | 1192 | // Destroy no registers! |
duke@435 | 1193 | // |
iveresov@2606 | 1194 | void gen_write_ref_array_pre_barrier(Register addr, Register count, bool dest_uninitialized) { |
duke@435 | 1195 | BarrierSet* bs = Universe::heap()->barrier_set(); |
duke@435 | 1196 | switch (bs->kind()) { |
duke@435 | 1197 | case BarrierSet::G1SATBCT: |
duke@435 | 1198 | case BarrierSet::G1SATBCTLogging: |
iveresov@2606 | 1199 | // With G1, don't generate the call if we statically know that the target in uninitialized |
iveresov@2606 | 1200 | if (!dest_uninitialized) { |
iveresov@2606 | 1201 | __ pusha(); // push registers |
iveresov@2606 | 1202 | if (count == c_rarg0) { |
iveresov@2606 | 1203 | if (addr == c_rarg1) { |
iveresov@2606 | 1204 | // exactly backwards!! |
iveresov@2606 | 1205 | __ xchgptr(c_rarg1, c_rarg0); |
iveresov@2606 | 1206 | } else { |
iveresov@2606 | 1207 | __ movptr(c_rarg1, count); |
iveresov@2606 | 1208 | __ movptr(c_rarg0, addr); |
iveresov@2606 | 1209 | } |
iveresov@2606 | 1210 | } else { |
iveresov@2606 | 1211 | __ movptr(c_rarg0, addr); |
iveresov@2606 | 1212 | __ movptr(c_rarg1, count); |
iveresov@2606 | 1213 | } |
iveresov@2606 | 1214 | __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2); |
iveresov@2606 | 1215 | __ popa(); |
duke@435 | 1216 | } |
iveresov@2606 | 1217 | break; |
duke@435 | 1218 | case BarrierSet::CardTableModRef: |
duke@435 | 1219 | case BarrierSet::CardTableExtension: |
duke@435 | 1220 | case BarrierSet::ModRef: |
duke@435 | 1221 | break; |
ysr@777 | 1222 | default: |
duke@435 | 1223 | ShouldNotReachHere(); |
duke@435 | 1224 | |
duke@435 | 1225 | } |
duke@435 | 1226 | } |
duke@435 | 1227 | |
duke@435 | 1228 | // |
duke@435 | 1229 | // Generate code for an array write post barrier |
duke@435 | 1230 | // |
duke@435 | 1231 | // Input: |
duke@435 | 1232 | // start - register containing starting address of destination array |
duke@435 | 1233 | // end - register containing ending address of destination array |
duke@435 | 1234 | // scratch - scratch register |
duke@435 | 1235 | // |
duke@435 | 1236 | // The input registers are overwritten. |
duke@435 | 1237 | // The ending address is inclusive. |
duke@435 | 1238 | void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) { |
duke@435 | 1239 | assert_different_registers(start, end, scratch); |
duke@435 | 1240 | BarrierSet* bs = Universe::heap()->barrier_set(); |
duke@435 | 1241 | switch (bs->kind()) { |
duke@435 | 1242 | case BarrierSet::G1SATBCT: |
duke@435 | 1243 | case BarrierSet::G1SATBCTLogging: |
duke@435 | 1244 | |
duke@435 | 1245 | { |
never@739 | 1246 | __ pusha(); // push registers (overkill) |
duke@435 | 1247 | // must compute element count unless barrier set interface is changed (other platforms supply count) |
duke@435 | 1248 | assert_different_registers(start, end, scratch); |
ysr@1280 | 1249 | __ lea(scratch, Address(end, BytesPerHeapOop)); |
ysr@1280 | 1250 | __ subptr(scratch, start); // subtract start to get #bytes |
ysr@1280 | 1251 | __ shrptr(scratch, LogBytesPerHeapOop); // convert to element count |
never@739 | 1252 | __ mov(c_rarg0, start); |
never@739 | 1253 | __ mov(c_rarg1, scratch); |
apetrusenko@1627 | 1254 | __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2); |
never@739 | 1255 | __ popa(); |
duke@435 | 1256 | } |
duke@435 | 1257 | break; |
duke@435 | 1258 | case BarrierSet::CardTableModRef: |
duke@435 | 1259 | case BarrierSet::CardTableExtension: |
duke@435 | 1260 | { |
duke@435 | 1261 | CardTableModRefBS* ct = (CardTableModRefBS*)bs; |
duke@435 | 1262 | assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); |
duke@435 | 1263 | |
duke@435 | 1264 | Label L_loop; |
duke@435 | 1265 | |
never@739 | 1266 | __ shrptr(start, CardTableModRefBS::card_shift); |
ysr@1280 | 1267 | __ addptr(end, BytesPerHeapOop); |
never@739 | 1268 | __ shrptr(end, CardTableModRefBS::card_shift); |
never@739 | 1269 | __ subptr(end, start); // number of bytes to copy |
duke@435 | 1270 | |
never@684 | 1271 | intptr_t disp = (intptr_t) ct->byte_map_base; |
twisti@3310 | 1272 | if (Assembler::is_simm32(disp)) { |
never@684 | 1273 | Address cardtable(noreg, noreg, Address::no_scale, disp); |
never@684 | 1274 | __ lea(scratch, cardtable); |
never@684 | 1275 | } else { |
never@684 | 1276 | ExternalAddress cardtable((address)disp); |
never@684 | 1277 | __ lea(scratch, cardtable); |
never@684 | 1278 | } |
never@684 | 1279 | |
duke@435 | 1280 | const Register count = end; // 'end' register contains bytes count now |
never@739 | 1281 | __ addptr(start, scratch); |
duke@435 | 1282 | __ BIND(L_loop); |
duke@435 | 1283 | __ movb(Address(start, count, Address::times_1), 0); |
never@739 | 1284 | __ decrement(count); |
duke@435 | 1285 | __ jcc(Assembler::greaterEqual, L_loop); |
duke@435 | 1286 | } |
ysr@777 | 1287 | break; |
ysr@777 | 1288 | default: |
ysr@777 | 1289 | ShouldNotReachHere(); |
ysr@777 | 1290 | |
ysr@777 | 1291 | } |
ysr@777 | 1292 | } |
duke@435 | 1293 | |
kvn@840 | 1294 | |
duke@435 | 1295 | // Copy big chunks forward |
duke@435 | 1296 | // |
duke@435 | 1297 | // Inputs: |
duke@435 | 1298 | // end_from - source arrays end address |
duke@435 | 1299 | // end_to - destination array end address |
duke@435 | 1300 | // qword_count - 64-bits element count, negative |
duke@435 | 1301 | // to - scratch |
duke@435 | 1302 | // L_copy_32_bytes - entry label |
duke@435 | 1303 | // L_copy_8_bytes - exit label |
duke@435 | 1304 | // |
duke@435 | 1305 | void copy_32_bytes_forward(Register end_from, Register end_to, |
duke@435 | 1306 | Register qword_count, Register to, |
duke@435 | 1307 | Label& L_copy_32_bytes, Label& L_copy_8_bytes) { |
duke@435 | 1308 | DEBUG_ONLY(__ stop("enter at entry label, not here")); |
duke@435 | 1309 | Label L_loop; |
kvn@1800 | 1310 | __ align(OptoLoopAlignment); |
duke@435 | 1311 | __ BIND(L_loop); |
kvn@840 | 1312 | if(UseUnalignedLoadStores) { |
kvn@840 | 1313 | __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24)); |
kvn@840 | 1314 | __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0); |
kvn@840 | 1315 | __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8)); |
kvn@840 | 1316 | __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1); |
kvn@840 | 1317 | |
kvn@840 | 1318 | } else { |
kvn@840 | 1319 | __ movq(to, Address(end_from, qword_count, Address::times_8, -24)); |
kvn@840 | 1320 | __ movq(Address(end_to, qword_count, Address::times_8, -24), to); |
kvn@840 | 1321 | __ movq(to, Address(end_from, qword_count, Address::times_8, -16)); |
kvn@840 | 1322 | __ movq(Address(end_to, qword_count, Address::times_8, -16), to); |
kvn@840 | 1323 | __ movq(to, Address(end_from, qword_count, Address::times_8, - 8)); |
kvn@840 | 1324 | __ movq(Address(end_to, qword_count, Address::times_8, - 8), to); |
kvn@840 | 1325 | __ movq(to, Address(end_from, qword_count, Address::times_8, - 0)); |
kvn@840 | 1326 | __ movq(Address(end_to, qword_count, Address::times_8, - 0), to); |
kvn@840 | 1327 | } |
duke@435 | 1328 | __ BIND(L_copy_32_bytes); |
never@739 | 1329 | __ addptr(qword_count, 4); |
duke@435 | 1330 | __ jcc(Assembler::lessEqual, L_loop); |
never@739 | 1331 | __ subptr(qword_count, 4); |
duke@435 | 1332 | __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords |
duke@435 | 1333 | } |
duke@435 | 1334 | |
duke@435 | 1335 | |
duke@435 | 1336 | // Copy big chunks backward |
duke@435 | 1337 | // |
duke@435 | 1338 | // Inputs: |
duke@435 | 1339 | // from - source arrays address |
duke@435 | 1340 | // dest - destination array address |
duke@435 | 1341 | // qword_count - 64-bits element count |
duke@435 | 1342 | // to - scratch |
duke@435 | 1343 | // L_copy_32_bytes - entry label |
duke@435 | 1344 | // L_copy_8_bytes - exit label |
duke@435 | 1345 | // |
duke@435 | 1346 | void copy_32_bytes_backward(Register from, Register dest, |
duke@435 | 1347 | Register qword_count, Register to, |
duke@435 | 1348 | Label& L_copy_32_bytes, Label& L_copy_8_bytes) { |
duke@435 | 1349 | DEBUG_ONLY(__ stop("enter at entry label, not here")); |
duke@435 | 1350 | Label L_loop; |
kvn@1800 | 1351 | __ align(OptoLoopAlignment); |
duke@435 | 1352 | __ BIND(L_loop); |
kvn@840 | 1353 | if(UseUnalignedLoadStores) { |
kvn@840 | 1354 | __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16)); |
kvn@840 | 1355 | __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0); |
kvn@840 | 1356 | __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0)); |
kvn@840 | 1357 | __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1); |
kvn@840 | 1358 | |
kvn@840 | 1359 | } else { |
kvn@840 | 1360 | __ movq(to, Address(from, qword_count, Address::times_8, 24)); |
kvn@840 | 1361 | __ movq(Address(dest, qword_count, Address::times_8, 24), to); |
kvn@840 | 1362 | __ movq(to, Address(from, qword_count, Address::times_8, 16)); |
kvn@840 | 1363 | __ movq(Address(dest, qword_count, Address::times_8, 16), to); |
kvn@840 | 1364 | __ movq(to, Address(from, qword_count, Address::times_8, 8)); |
kvn@840 | 1365 | __ movq(Address(dest, qword_count, Address::times_8, 8), to); |
kvn@840 | 1366 | __ movq(to, Address(from, qword_count, Address::times_8, 0)); |
kvn@840 | 1367 | __ movq(Address(dest, qword_count, Address::times_8, 0), to); |
kvn@840 | 1368 | } |
duke@435 | 1369 | __ BIND(L_copy_32_bytes); |
never@739 | 1370 | __ subptr(qword_count, 4); |
duke@435 | 1371 | __ jcc(Assembler::greaterEqual, L_loop); |
never@739 | 1372 | __ addptr(qword_count, 4); |
duke@435 | 1373 | __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords |
duke@435 | 1374 | } |
duke@435 | 1375 | |
duke@435 | 1376 | |
duke@435 | 1377 | // Arguments: |
duke@435 | 1378 | // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary |
duke@435 | 1379 | // ignored |
duke@435 | 1380 | // name - stub name string |
duke@435 | 1381 | // |
duke@435 | 1382 | // Inputs: |
duke@435 | 1383 | // c_rarg0 - source array address |
duke@435 | 1384 | // c_rarg1 - destination array address |
duke@435 | 1385 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1386 | // |
duke@435 | 1387 | // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries, |
duke@435 | 1388 | // we let the hardware handle it. The one to eight bytes within words, |
duke@435 | 1389 | // dwords or qwords that span cache line boundaries will still be loaded |
duke@435 | 1390 | // and stored atomically. |
duke@435 | 1391 | // |
duke@435 | 1392 | // Side Effects: |
duke@435 | 1393 | // disjoint_byte_copy_entry is set to the no-overlap entry point |
duke@435 | 1394 | // used by generate_conjoint_byte_copy(). |
duke@435 | 1395 | // |
iveresov@2595 | 1396 | address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) { |
duke@435 | 1397 | __ align(CodeEntryAlignment); |
duke@435 | 1398 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1399 | address start = __ pc(); |
duke@435 | 1400 | |
duke@435 | 1401 | Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes; |
duke@435 | 1402 | Label L_copy_byte, L_exit; |
duke@435 | 1403 | const Register from = rdi; // source array address |
duke@435 | 1404 | const Register to = rsi; // destination array address |
duke@435 | 1405 | const Register count = rdx; // elements count |
duke@435 | 1406 | const Register byte_count = rcx; |
duke@435 | 1407 | const Register qword_count = count; |
duke@435 | 1408 | const Register end_from = from; // source array end address |
duke@435 | 1409 | const Register end_to = to; // destination array end address |
duke@435 | 1410 | // End pointers are inclusive, and if count is not zero they point |
duke@435 | 1411 | // to the last unit copied: end_to[0] := end_from[0] |
duke@435 | 1412 | |
duke@435 | 1413 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1414 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 1415 | |
iveresov@2595 | 1416 | if (entry != NULL) { |
iveresov@2595 | 1417 | *entry = __ pc(); |
iveresov@2595 | 1418 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1419 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1420 | } |
duke@435 | 1421 | |
duke@435 | 1422 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 1423 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 1424 | |
duke@435 | 1425 | // 'from', 'to' and 'count' are now valid |
never@739 | 1426 | __ movptr(byte_count, count); |
never@739 | 1427 | __ shrptr(count, 3); // count => qword_count |
duke@435 | 1428 | |
duke@435 | 1429 | // Copy from low to high addresses. Use 'to' as scratch. |
never@739 | 1430 | __ lea(end_from, Address(from, qword_count, Address::times_8, -8)); |
never@739 | 1431 | __ lea(end_to, Address(to, qword_count, Address::times_8, -8)); |
never@739 | 1432 | __ negptr(qword_count); // make the count negative |
duke@435 | 1433 | __ jmp(L_copy_32_bytes); |
duke@435 | 1434 | |
duke@435 | 1435 | // Copy trailing qwords |
duke@435 | 1436 | __ BIND(L_copy_8_bytes); |
duke@435 | 1437 | __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); |
duke@435 | 1438 | __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); |
never@739 | 1439 | __ increment(qword_count); |
duke@435 | 1440 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 1441 | |
duke@435 | 1442 | // Check for and copy trailing dword |
duke@435 | 1443 | __ BIND(L_copy_4_bytes); |
never@739 | 1444 | __ testl(byte_count, 4); |
duke@435 | 1445 | __ jccb(Assembler::zero, L_copy_2_bytes); |
duke@435 | 1446 | __ movl(rax, Address(end_from, 8)); |
duke@435 | 1447 | __ movl(Address(end_to, 8), rax); |
duke@435 | 1448 | |
never@739 | 1449 | __ addptr(end_from, 4); |
never@739 | 1450 | __ addptr(end_to, 4); |
duke@435 | 1451 | |
duke@435 | 1452 | // Check for and copy trailing word |
duke@435 | 1453 | __ BIND(L_copy_2_bytes); |
never@739 | 1454 | __ testl(byte_count, 2); |
duke@435 | 1455 | __ jccb(Assembler::zero, L_copy_byte); |
duke@435 | 1456 | __ movw(rax, Address(end_from, 8)); |
duke@435 | 1457 | __ movw(Address(end_to, 8), rax); |
duke@435 | 1458 | |
never@739 | 1459 | __ addptr(end_from, 2); |
never@739 | 1460 | __ addptr(end_to, 2); |
duke@435 | 1461 | |
duke@435 | 1462 | // Check for and copy trailing byte |
duke@435 | 1463 | __ BIND(L_copy_byte); |
never@739 | 1464 | __ testl(byte_count, 1); |
duke@435 | 1465 | __ jccb(Assembler::zero, L_exit); |
duke@435 | 1466 | __ movb(rax, Address(end_from, 8)); |
duke@435 | 1467 | __ movb(Address(end_to, 8), rax); |
duke@435 | 1468 | |
duke@435 | 1469 | __ BIND(L_exit); |
duke@435 | 1470 | restore_arg_regs(); |
never@3314 | 1471 | inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1472 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1473 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1474 | __ ret(0); |
duke@435 | 1475 | |
duke@435 | 1476 | // Copy in 32-bytes chunks |
duke@435 | 1477 | copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); |
duke@435 | 1478 | __ jmp(L_copy_4_bytes); |
duke@435 | 1479 | |
duke@435 | 1480 | return start; |
duke@435 | 1481 | } |
duke@435 | 1482 | |
duke@435 | 1483 | // Arguments: |
duke@435 | 1484 | // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary |
duke@435 | 1485 | // ignored |
duke@435 | 1486 | // name - stub name string |
duke@435 | 1487 | // |
duke@435 | 1488 | // Inputs: |
duke@435 | 1489 | // c_rarg0 - source array address |
duke@435 | 1490 | // c_rarg1 - destination array address |
duke@435 | 1491 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1492 | // |
duke@435 | 1493 | // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries, |
duke@435 | 1494 | // we let the hardware handle it. The one to eight bytes within words, |
duke@435 | 1495 | // dwords or qwords that span cache line boundaries will still be loaded |
duke@435 | 1496 | // and stored atomically. |
duke@435 | 1497 | // |
iveresov@2595 | 1498 | address generate_conjoint_byte_copy(bool aligned, address nooverlap_target, |
iveresov@2595 | 1499 | address* entry, const char *name) { |
duke@435 | 1500 | __ align(CodeEntryAlignment); |
duke@435 | 1501 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1502 | address start = __ pc(); |
duke@435 | 1503 | |
duke@435 | 1504 | Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes; |
duke@435 | 1505 | const Register from = rdi; // source array address |
duke@435 | 1506 | const Register to = rsi; // destination array address |
duke@435 | 1507 | const Register count = rdx; // elements count |
duke@435 | 1508 | const Register byte_count = rcx; |
duke@435 | 1509 | const Register qword_count = count; |
duke@435 | 1510 | |
duke@435 | 1511 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1512 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 1513 | |
iveresov@2595 | 1514 | if (entry != NULL) { |
iveresov@2595 | 1515 | *entry = __ pc(); |
iveresov@2595 | 1516 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1517 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1518 | } |
iveresov@2595 | 1519 | |
iveresov@2595 | 1520 | array_overlap_test(nooverlap_target, Address::times_1); |
duke@435 | 1521 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 1522 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 1523 | |
duke@435 | 1524 | // 'from', 'to' and 'count' are now valid |
never@739 | 1525 | __ movptr(byte_count, count); |
never@739 | 1526 | __ shrptr(count, 3); // count => qword_count |
duke@435 | 1527 | |
duke@435 | 1528 | // Copy from high to low addresses. |
duke@435 | 1529 | |
duke@435 | 1530 | // Check for and copy trailing byte |
never@739 | 1531 | __ testl(byte_count, 1); |
duke@435 | 1532 | __ jcc(Assembler::zero, L_copy_2_bytes); |
duke@435 | 1533 | __ movb(rax, Address(from, byte_count, Address::times_1, -1)); |
duke@435 | 1534 | __ movb(Address(to, byte_count, Address::times_1, -1), rax); |
never@739 | 1535 | __ decrement(byte_count); // Adjust for possible trailing word |
duke@435 | 1536 | |
duke@435 | 1537 | // Check for and copy trailing word |
duke@435 | 1538 | __ BIND(L_copy_2_bytes); |
never@739 | 1539 | __ testl(byte_count, 2); |
duke@435 | 1540 | __ jcc(Assembler::zero, L_copy_4_bytes); |
duke@435 | 1541 | __ movw(rax, Address(from, byte_count, Address::times_1, -2)); |
duke@435 | 1542 | __ movw(Address(to, byte_count, Address::times_1, -2), rax); |
duke@435 | 1543 | |
duke@435 | 1544 | // Check for and copy trailing dword |
duke@435 | 1545 | __ BIND(L_copy_4_bytes); |
never@739 | 1546 | __ testl(byte_count, 4); |
duke@435 | 1547 | __ jcc(Assembler::zero, L_copy_32_bytes); |
duke@435 | 1548 | __ movl(rax, Address(from, qword_count, Address::times_8)); |
duke@435 | 1549 | __ movl(Address(to, qword_count, Address::times_8), rax); |
duke@435 | 1550 | __ jmp(L_copy_32_bytes); |
duke@435 | 1551 | |
duke@435 | 1552 | // Copy trailing qwords |
duke@435 | 1553 | __ BIND(L_copy_8_bytes); |
duke@435 | 1554 | __ movq(rax, Address(from, qword_count, Address::times_8, -8)); |
duke@435 | 1555 | __ movq(Address(to, qword_count, Address::times_8, -8), rax); |
never@739 | 1556 | __ decrement(qword_count); |
duke@435 | 1557 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 1558 | |
duke@435 | 1559 | restore_arg_regs(); |
never@3314 | 1560 | inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1561 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1562 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1563 | __ ret(0); |
duke@435 | 1564 | |
duke@435 | 1565 | // Copy in 32-bytes chunks |
duke@435 | 1566 | copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); |
duke@435 | 1567 | |
duke@435 | 1568 | restore_arg_regs(); |
never@3314 | 1569 | inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1570 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1571 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1572 | __ ret(0); |
duke@435 | 1573 | |
duke@435 | 1574 | return start; |
duke@435 | 1575 | } |
duke@435 | 1576 | |
duke@435 | 1577 | // Arguments: |
duke@435 | 1578 | // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary |
duke@435 | 1579 | // ignored |
duke@435 | 1580 | // name - stub name string |
duke@435 | 1581 | // |
duke@435 | 1582 | // Inputs: |
duke@435 | 1583 | // c_rarg0 - source array address |
duke@435 | 1584 | // c_rarg1 - destination array address |
duke@435 | 1585 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1586 | // |
duke@435 | 1587 | // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we |
duke@435 | 1588 | // let the hardware handle it. The two or four words within dwords |
duke@435 | 1589 | // or qwords that span cache line boundaries will still be loaded |
duke@435 | 1590 | // and stored atomically. |
duke@435 | 1591 | // |
duke@435 | 1592 | // Side Effects: |
duke@435 | 1593 | // disjoint_short_copy_entry is set to the no-overlap entry point |
duke@435 | 1594 | // used by generate_conjoint_short_copy(). |
duke@435 | 1595 | // |
iveresov@2595 | 1596 | address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) { |
duke@435 | 1597 | __ align(CodeEntryAlignment); |
duke@435 | 1598 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1599 | address start = __ pc(); |
duke@435 | 1600 | |
duke@435 | 1601 | Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit; |
duke@435 | 1602 | const Register from = rdi; // source array address |
duke@435 | 1603 | const Register to = rsi; // destination array address |
duke@435 | 1604 | const Register count = rdx; // elements count |
duke@435 | 1605 | const Register word_count = rcx; |
duke@435 | 1606 | const Register qword_count = count; |
duke@435 | 1607 | const Register end_from = from; // source array end address |
duke@435 | 1608 | const Register end_to = to; // destination array end address |
duke@435 | 1609 | // End pointers are inclusive, and if count is not zero they point |
duke@435 | 1610 | // to the last unit copied: end_to[0] := end_from[0] |
duke@435 | 1611 | |
duke@435 | 1612 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1613 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 1614 | |
iveresov@2595 | 1615 | if (entry != NULL) { |
iveresov@2595 | 1616 | *entry = __ pc(); |
iveresov@2595 | 1617 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1618 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1619 | } |
duke@435 | 1620 | |
duke@435 | 1621 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 1622 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 1623 | |
duke@435 | 1624 | // 'from', 'to' and 'count' are now valid |
never@739 | 1625 | __ movptr(word_count, count); |
never@739 | 1626 | __ shrptr(count, 2); // count => qword_count |
duke@435 | 1627 | |
duke@435 | 1628 | // Copy from low to high addresses. Use 'to' as scratch. |
never@739 | 1629 | __ lea(end_from, Address(from, qword_count, Address::times_8, -8)); |
never@739 | 1630 | __ lea(end_to, Address(to, qword_count, Address::times_8, -8)); |
never@739 | 1631 | __ negptr(qword_count); |
duke@435 | 1632 | __ jmp(L_copy_32_bytes); |
duke@435 | 1633 | |
duke@435 | 1634 | // Copy trailing qwords |
duke@435 | 1635 | __ BIND(L_copy_8_bytes); |
duke@435 | 1636 | __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); |
duke@435 | 1637 | __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); |
never@739 | 1638 | __ increment(qword_count); |
duke@435 | 1639 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 1640 | |
duke@435 | 1641 | // Original 'dest' is trashed, so we can't use it as a |
duke@435 | 1642 | // base register for a possible trailing word copy |
duke@435 | 1643 | |
duke@435 | 1644 | // Check for and copy trailing dword |
duke@435 | 1645 | __ BIND(L_copy_4_bytes); |
never@739 | 1646 | __ testl(word_count, 2); |
duke@435 | 1647 | __ jccb(Assembler::zero, L_copy_2_bytes); |
duke@435 | 1648 | __ movl(rax, Address(end_from, 8)); |
duke@435 | 1649 | __ movl(Address(end_to, 8), rax); |
duke@435 | 1650 | |
never@739 | 1651 | __ addptr(end_from, 4); |
never@739 | 1652 | __ addptr(end_to, 4); |
duke@435 | 1653 | |
duke@435 | 1654 | // Check for and copy trailing word |
duke@435 | 1655 | __ BIND(L_copy_2_bytes); |
never@739 | 1656 | __ testl(word_count, 1); |
duke@435 | 1657 | __ jccb(Assembler::zero, L_exit); |
duke@435 | 1658 | __ movw(rax, Address(end_from, 8)); |
duke@435 | 1659 | __ movw(Address(end_to, 8), rax); |
duke@435 | 1660 | |
duke@435 | 1661 | __ BIND(L_exit); |
duke@435 | 1662 | restore_arg_regs(); |
never@3314 | 1663 | inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1664 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1665 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1666 | __ ret(0); |
duke@435 | 1667 | |
duke@435 | 1668 | // Copy in 32-bytes chunks |
duke@435 | 1669 | copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); |
duke@435 | 1670 | __ jmp(L_copy_4_bytes); |
duke@435 | 1671 | |
duke@435 | 1672 | return start; |
duke@435 | 1673 | } |
duke@435 | 1674 | |
never@2118 | 1675 | address generate_fill(BasicType t, bool aligned, const char *name) { |
never@2118 | 1676 | __ align(CodeEntryAlignment); |
never@2118 | 1677 | StubCodeMark mark(this, "StubRoutines", name); |
never@2118 | 1678 | address start = __ pc(); |
never@2118 | 1679 | |
never@2118 | 1680 | BLOCK_COMMENT("Entry:"); |
never@2118 | 1681 | |
never@2118 | 1682 | const Register to = c_rarg0; // source array address |
never@2118 | 1683 | const Register value = c_rarg1; // value |
never@2118 | 1684 | const Register count = c_rarg2; // elements count |
never@2118 | 1685 | |
never@2118 | 1686 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
never@2118 | 1687 | |
never@2118 | 1688 | __ generate_fill(t, aligned, to, value, count, rax, xmm0); |
never@2118 | 1689 | |
never@2118 | 1690 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
never@2118 | 1691 | __ ret(0); |
never@2118 | 1692 | return start; |
never@2118 | 1693 | } |
never@2118 | 1694 | |
duke@435 | 1695 | // Arguments: |
duke@435 | 1696 | // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary |
duke@435 | 1697 | // ignored |
duke@435 | 1698 | // name - stub name string |
duke@435 | 1699 | // |
duke@435 | 1700 | // Inputs: |
duke@435 | 1701 | // c_rarg0 - source array address |
duke@435 | 1702 | // c_rarg1 - destination array address |
duke@435 | 1703 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1704 | // |
duke@435 | 1705 | // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we |
duke@435 | 1706 | // let the hardware handle it. The two or four words within dwords |
duke@435 | 1707 | // or qwords that span cache line boundaries will still be loaded |
duke@435 | 1708 | // and stored atomically. |
duke@435 | 1709 | // |
iveresov@2595 | 1710 | address generate_conjoint_short_copy(bool aligned, address nooverlap_target, |
iveresov@2595 | 1711 | address *entry, const char *name) { |
duke@435 | 1712 | __ align(CodeEntryAlignment); |
duke@435 | 1713 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1714 | address start = __ pc(); |
duke@435 | 1715 | |
duke@435 | 1716 | Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes; |
duke@435 | 1717 | const Register from = rdi; // source array address |
duke@435 | 1718 | const Register to = rsi; // destination array address |
duke@435 | 1719 | const Register count = rdx; // elements count |
duke@435 | 1720 | const Register word_count = rcx; |
duke@435 | 1721 | const Register qword_count = count; |
duke@435 | 1722 | |
duke@435 | 1723 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1724 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 1725 | |
iveresov@2595 | 1726 | if (entry != NULL) { |
iveresov@2595 | 1727 | *entry = __ pc(); |
iveresov@2595 | 1728 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1729 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1730 | } |
iveresov@2595 | 1731 | |
iveresov@2595 | 1732 | array_overlap_test(nooverlap_target, Address::times_2); |
duke@435 | 1733 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 1734 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 1735 | |
duke@435 | 1736 | // 'from', 'to' and 'count' are now valid |
never@739 | 1737 | __ movptr(word_count, count); |
never@739 | 1738 | __ shrptr(count, 2); // count => qword_count |
duke@435 | 1739 | |
duke@435 | 1740 | // Copy from high to low addresses. Use 'to' as scratch. |
duke@435 | 1741 | |
duke@435 | 1742 | // Check for and copy trailing word |
never@739 | 1743 | __ testl(word_count, 1); |
duke@435 | 1744 | __ jccb(Assembler::zero, L_copy_4_bytes); |
duke@435 | 1745 | __ movw(rax, Address(from, word_count, Address::times_2, -2)); |
duke@435 | 1746 | __ movw(Address(to, word_count, Address::times_2, -2), rax); |
duke@435 | 1747 | |
duke@435 | 1748 | // Check for and copy trailing dword |
duke@435 | 1749 | __ BIND(L_copy_4_bytes); |
never@739 | 1750 | __ testl(word_count, 2); |
duke@435 | 1751 | __ jcc(Assembler::zero, L_copy_32_bytes); |
duke@435 | 1752 | __ movl(rax, Address(from, qword_count, Address::times_8)); |
duke@435 | 1753 | __ movl(Address(to, qword_count, Address::times_8), rax); |
duke@435 | 1754 | __ jmp(L_copy_32_bytes); |
duke@435 | 1755 | |
duke@435 | 1756 | // Copy trailing qwords |
duke@435 | 1757 | __ BIND(L_copy_8_bytes); |
duke@435 | 1758 | __ movq(rax, Address(from, qword_count, Address::times_8, -8)); |
duke@435 | 1759 | __ movq(Address(to, qword_count, Address::times_8, -8), rax); |
never@739 | 1760 | __ decrement(qword_count); |
duke@435 | 1761 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 1762 | |
duke@435 | 1763 | restore_arg_regs(); |
never@3314 | 1764 | inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1765 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1766 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1767 | __ ret(0); |
duke@435 | 1768 | |
duke@435 | 1769 | // Copy in 32-bytes chunks |
duke@435 | 1770 | copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); |
duke@435 | 1771 | |
duke@435 | 1772 | restore_arg_regs(); |
never@3314 | 1773 | inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1774 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1775 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1776 | __ ret(0); |
duke@435 | 1777 | |
duke@435 | 1778 | return start; |
duke@435 | 1779 | } |
duke@435 | 1780 | |
duke@435 | 1781 | // Arguments: |
duke@435 | 1782 | // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary |
duke@435 | 1783 | // ignored |
coleenp@548 | 1784 | // is_oop - true => oop array, so generate store check code |
duke@435 | 1785 | // name - stub name string |
duke@435 | 1786 | // |
duke@435 | 1787 | // Inputs: |
duke@435 | 1788 | // c_rarg0 - source array address |
duke@435 | 1789 | // c_rarg1 - destination array address |
duke@435 | 1790 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1791 | // |
duke@435 | 1792 | // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let |
duke@435 | 1793 | // the hardware handle it. The two dwords within qwords that span |
duke@435 | 1794 | // cache line boundaries will still be loaded and stored atomicly. |
duke@435 | 1795 | // |
duke@435 | 1796 | // Side Effects: |
duke@435 | 1797 | // disjoint_int_copy_entry is set to the no-overlap entry point |
coleenp@548 | 1798 | // used by generate_conjoint_int_oop_copy(). |
duke@435 | 1799 | // |
iveresov@2606 | 1800 | address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry, |
iveresov@2606 | 1801 | const char *name, bool dest_uninitialized = false) { |
duke@435 | 1802 | __ align(CodeEntryAlignment); |
duke@435 | 1803 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1804 | address start = __ pc(); |
duke@435 | 1805 | |
duke@435 | 1806 | Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit; |
duke@435 | 1807 | const Register from = rdi; // source array address |
duke@435 | 1808 | const Register to = rsi; // destination array address |
duke@435 | 1809 | const Register count = rdx; // elements count |
duke@435 | 1810 | const Register dword_count = rcx; |
duke@435 | 1811 | const Register qword_count = count; |
duke@435 | 1812 | const Register end_from = from; // source array end address |
duke@435 | 1813 | const Register end_to = to; // destination array end address |
coleenp@548 | 1814 | const Register saved_to = r11; // saved destination array address |
duke@435 | 1815 | // End pointers are inclusive, and if count is not zero they point |
duke@435 | 1816 | // to the last unit copied: end_to[0] := end_from[0] |
duke@435 | 1817 | |
duke@435 | 1818 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1819 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 1820 | |
iveresov@2595 | 1821 | if (entry != NULL) { |
iveresov@2595 | 1822 | *entry = __ pc(); |
iveresov@2595 | 1823 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1824 | BLOCK_COMMENT("Entry:"); |
coleenp@548 | 1825 | } |
coleenp@548 | 1826 | |
duke@435 | 1827 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 1828 | // r9 and r10 may be used to save non-volatile registers |
coleenp@548 | 1829 | if (is_oop) { |
coleenp@548 | 1830 | __ movq(saved_to, to); |
iveresov@2606 | 1831 | gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); |
coleenp@548 | 1832 | } |
coleenp@548 | 1833 | |
duke@435 | 1834 | // 'from', 'to' and 'count' are now valid |
never@739 | 1835 | __ movptr(dword_count, count); |
never@739 | 1836 | __ shrptr(count, 1); // count => qword_count |
duke@435 | 1837 | |
duke@435 | 1838 | // Copy from low to high addresses. Use 'to' as scratch. |
never@739 | 1839 | __ lea(end_from, Address(from, qword_count, Address::times_8, -8)); |
never@739 | 1840 | __ lea(end_to, Address(to, qword_count, Address::times_8, -8)); |
never@739 | 1841 | __ negptr(qword_count); |
duke@435 | 1842 | __ jmp(L_copy_32_bytes); |
duke@435 | 1843 | |
duke@435 | 1844 | // Copy trailing qwords |
duke@435 | 1845 | __ BIND(L_copy_8_bytes); |
duke@435 | 1846 | __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); |
duke@435 | 1847 | __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); |
never@739 | 1848 | __ increment(qword_count); |
duke@435 | 1849 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 1850 | |
duke@435 | 1851 | // Check for and copy trailing dword |
duke@435 | 1852 | __ BIND(L_copy_4_bytes); |
never@739 | 1853 | __ testl(dword_count, 1); // Only byte test since the value is 0 or 1 |
duke@435 | 1854 | __ jccb(Assembler::zero, L_exit); |
duke@435 | 1855 | __ movl(rax, Address(end_from, 8)); |
duke@435 | 1856 | __ movl(Address(end_to, 8), rax); |
duke@435 | 1857 | |
duke@435 | 1858 | __ BIND(L_exit); |
coleenp@548 | 1859 | if (is_oop) { |
coleenp@548 | 1860 | __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4)); |
coleenp@548 | 1861 | gen_write_ref_array_post_barrier(saved_to, end_to, rax); |
coleenp@548 | 1862 | } |
duke@435 | 1863 | restore_arg_regs(); |
never@3314 | 1864 | inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1865 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1866 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1867 | __ ret(0); |
duke@435 | 1868 | |
duke@435 | 1869 | // Copy 32-bytes chunks |
duke@435 | 1870 | copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); |
duke@435 | 1871 | __ jmp(L_copy_4_bytes); |
duke@435 | 1872 | |
duke@435 | 1873 | return start; |
duke@435 | 1874 | } |
duke@435 | 1875 | |
duke@435 | 1876 | // Arguments: |
duke@435 | 1877 | // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary |
duke@435 | 1878 | // ignored |
coleenp@548 | 1879 | // is_oop - true => oop array, so generate store check code |
duke@435 | 1880 | // name - stub name string |
duke@435 | 1881 | // |
duke@435 | 1882 | // Inputs: |
duke@435 | 1883 | // c_rarg0 - source array address |
duke@435 | 1884 | // c_rarg1 - destination array address |
duke@435 | 1885 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1886 | // |
duke@435 | 1887 | // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let |
duke@435 | 1888 | // the hardware handle it. The two dwords within qwords that span |
duke@435 | 1889 | // cache line boundaries will still be loaded and stored atomicly. |
duke@435 | 1890 | // |
iveresov@2595 | 1891 | address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target, |
iveresov@2606 | 1892 | address *entry, const char *name, |
iveresov@2606 | 1893 | bool dest_uninitialized = false) { |
duke@435 | 1894 | __ align(CodeEntryAlignment); |
duke@435 | 1895 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1896 | address start = __ pc(); |
duke@435 | 1897 | |
coleenp@548 | 1898 | Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit; |
duke@435 | 1899 | const Register from = rdi; // source array address |
duke@435 | 1900 | const Register to = rsi; // destination array address |
duke@435 | 1901 | const Register count = rdx; // elements count |
duke@435 | 1902 | const Register dword_count = rcx; |
duke@435 | 1903 | const Register qword_count = count; |
duke@435 | 1904 | |
duke@435 | 1905 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1906 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 1907 | |
iveresov@2595 | 1908 | if (entry != NULL) { |
iveresov@2595 | 1909 | *entry = __ pc(); |
iveresov@2595 | 1910 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1911 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1912 | } |
iveresov@2595 | 1913 | |
iveresov@2595 | 1914 | array_overlap_test(nooverlap_target, Address::times_4); |
iveresov@2595 | 1915 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
iveresov@2595 | 1916 | // r9 and r10 may be used to save non-volatile registers |
iveresov@2595 | 1917 | |
coleenp@548 | 1918 | if (is_oop) { |
coleenp@548 | 1919 | // no registers are destroyed by this call |
iveresov@2606 | 1920 | gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); |
coleenp@548 | 1921 | } |
coleenp@548 | 1922 | |
coleenp@548 | 1923 | assert_clean_int(count, rax); // Make sure 'count' is clean int. |
duke@435 | 1924 | // 'from', 'to' and 'count' are now valid |
never@739 | 1925 | __ movptr(dword_count, count); |
never@739 | 1926 | __ shrptr(count, 1); // count => qword_count |
duke@435 | 1927 | |
duke@435 | 1928 | // Copy from high to low addresses. Use 'to' as scratch. |
duke@435 | 1929 | |
duke@435 | 1930 | // Check for and copy trailing dword |
never@739 | 1931 | __ testl(dword_count, 1); |
duke@435 | 1932 | __ jcc(Assembler::zero, L_copy_32_bytes); |
duke@435 | 1933 | __ movl(rax, Address(from, dword_count, Address::times_4, -4)); |
duke@435 | 1934 | __ movl(Address(to, dword_count, Address::times_4, -4), rax); |
duke@435 | 1935 | __ jmp(L_copy_32_bytes); |
duke@435 | 1936 | |
duke@435 | 1937 | // Copy trailing qwords |
duke@435 | 1938 | __ BIND(L_copy_8_bytes); |
duke@435 | 1939 | __ movq(rax, Address(from, qword_count, Address::times_8, -8)); |
duke@435 | 1940 | __ movq(Address(to, qword_count, Address::times_8, -8), rax); |
never@739 | 1941 | __ decrement(qword_count); |
duke@435 | 1942 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 1943 | |
coleenp@548 | 1944 | if (is_oop) { |
coleenp@548 | 1945 | __ jmp(L_exit); |
coleenp@548 | 1946 | } |
duke@435 | 1947 | restore_arg_regs(); |
never@3314 | 1948 | inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1949 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1950 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1951 | __ ret(0); |
duke@435 | 1952 | |
duke@435 | 1953 | // Copy in 32-bytes chunks |
duke@435 | 1954 | copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); |
duke@435 | 1955 | |
coleenp@548 | 1956 | __ bind(L_exit); |
coleenp@548 | 1957 | if (is_oop) { |
coleenp@548 | 1958 | Register end_to = rdx; |
coleenp@548 | 1959 | __ leaq(end_to, Address(to, dword_count, Address::times_4, -4)); |
coleenp@548 | 1960 | gen_write_ref_array_post_barrier(to, end_to, rax); |
coleenp@548 | 1961 | } |
duke@435 | 1962 | restore_arg_regs(); |
never@3314 | 1963 | inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1964 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1965 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1966 | __ ret(0); |
duke@435 | 1967 | |
duke@435 | 1968 | return start; |
duke@435 | 1969 | } |
duke@435 | 1970 | |
duke@435 | 1971 | // Arguments: |
duke@435 | 1972 | // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes |
duke@435 | 1973 | // ignored |
duke@435 | 1974 | // is_oop - true => oop array, so generate store check code |
duke@435 | 1975 | // name - stub name string |
duke@435 | 1976 | // |
duke@435 | 1977 | // Inputs: |
duke@435 | 1978 | // c_rarg0 - source array address |
duke@435 | 1979 | // c_rarg1 - destination array address |
duke@435 | 1980 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1981 | // |
coleenp@548 | 1982 | // Side Effects: |
duke@435 | 1983 | // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the |
duke@435 | 1984 | // no-overlap entry point used by generate_conjoint_long_oop_copy(). |
duke@435 | 1985 | // |
iveresov@2606 | 1986 | address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry, |
iveresov@2606 | 1987 | const char *name, bool dest_uninitialized = false) { |
duke@435 | 1988 | __ align(CodeEntryAlignment); |
duke@435 | 1989 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1990 | address start = __ pc(); |
duke@435 | 1991 | |
duke@435 | 1992 | Label L_copy_32_bytes, L_copy_8_bytes, L_exit; |
duke@435 | 1993 | const Register from = rdi; // source array address |
duke@435 | 1994 | const Register to = rsi; // destination array address |
duke@435 | 1995 | const Register qword_count = rdx; // elements count |
duke@435 | 1996 | const Register end_from = from; // source array end address |
duke@435 | 1997 | const Register end_to = rcx; // destination array end address |
duke@435 | 1998 | const Register saved_to = to; |
duke@435 | 1999 | // End pointers are inclusive, and if count is not zero they point |
duke@435 | 2000 | // to the last unit copied: end_to[0] := end_from[0] |
duke@435 | 2001 | |
duke@435 | 2002 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2003 | // Save no-overlap entry point for generate_conjoint_long_oop_copy() |
duke@435 | 2004 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 2005 | |
iveresov@2595 | 2006 | if (entry != NULL) { |
iveresov@2595 | 2007 | *entry = __ pc(); |
iveresov@2595 | 2008 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 2009 | BLOCK_COMMENT("Entry:"); |
duke@435 | 2010 | } |
duke@435 | 2011 | |
duke@435 | 2012 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 2013 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 2014 | // 'from', 'to' and 'qword_count' are now valid |
iveresov@2595 | 2015 | if (is_oop) { |
iveresov@2595 | 2016 | // no registers are destroyed by this call |
iveresov@2606 | 2017 | gen_write_ref_array_pre_barrier(to, qword_count, dest_uninitialized); |
iveresov@2595 | 2018 | } |
duke@435 | 2019 | |
duke@435 | 2020 | // Copy from low to high addresses. Use 'to' as scratch. |
never@739 | 2021 | __ lea(end_from, Address(from, qword_count, Address::times_8, -8)); |
never@739 | 2022 | __ lea(end_to, Address(to, qword_count, Address::times_8, -8)); |
never@739 | 2023 | __ negptr(qword_count); |
duke@435 | 2024 | __ jmp(L_copy_32_bytes); |
duke@435 | 2025 | |
duke@435 | 2026 | // Copy trailing qwords |
duke@435 | 2027 | __ BIND(L_copy_8_bytes); |
duke@435 | 2028 | __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); |
duke@435 | 2029 | __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); |
never@739 | 2030 | __ increment(qword_count); |
duke@435 | 2031 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 2032 | |
duke@435 | 2033 | if (is_oop) { |
duke@435 | 2034 | __ jmp(L_exit); |
duke@435 | 2035 | } else { |
duke@435 | 2036 | restore_arg_regs(); |
never@3314 | 2037 | inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 2038 | __ xorptr(rax, rax); // return 0 |
duke@435 | 2039 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2040 | __ ret(0); |
duke@435 | 2041 | } |
duke@435 | 2042 | |
duke@435 | 2043 | // Copy 64-byte chunks |
duke@435 | 2044 | copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); |
duke@435 | 2045 | |
duke@435 | 2046 | if (is_oop) { |
duke@435 | 2047 | __ BIND(L_exit); |
duke@435 | 2048 | gen_write_ref_array_post_barrier(saved_to, end_to, rax); |
duke@435 | 2049 | } |
duke@435 | 2050 | restore_arg_regs(); |
never@3314 | 2051 | if (is_oop) { |
never@3314 | 2052 | inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free |
never@3314 | 2053 | } else { |
never@3314 | 2054 | inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free |
never@3314 | 2055 | } |
never@739 | 2056 | __ xorptr(rax, rax); // return 0 |
duke@435 | 2057 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2058 | __ ret(0); |
duke@435 | 2059 | |
duke@435 | 2060 | return start; |
duke@435 | 2061 | } |
duke@435 | 2062 | |
duke@435 | 2063 | // Arguments: |
duke@435 | 2064 | // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes |
duke@435 | 2065 | // ignored |
duke@435 | 2066 | // is_oop - true => oop array, so generate store check code |
duke@435 | 2067 | // name - stub name string |
duke@435 | 2068 | // |
duke@435 | 2069 | // Inputs: |
duke@435 | 2070 | // c_rarg0 - source array address |
duke@435 | 2071 | // c_rarg1 - destination array address |
duke@435 | 2072 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 2073 | // |
iveresov@2606 | 2074 | address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, |
iveresov@2606 | 2075 | address nooverlap_target, address *entry, |
iveresov@2606 | 2076 | const char *name, bool dest_uninitialized = false) { |
duke@435 | 2077 | __ align(CodeEntryAlignment); |
duke@435 | 2078 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2079 | address start = __ pc(); |
duke@435 | 2080 | |
duke@435 | 2081 | Label L_copy_32_bytes, L_copy_8_bytes, L_exit; |
duke@435 | 2082 | const Register from = rdi; // source array address |
duke@435 | 2083 | const Register to = rsi; // destination array address |
duke@435 | 2084 | const Register qword_count = rdx; // elements count |
duke@435 | 2085 | const Register saved_count = rcx; |
duke@435 | 2086 | |
duke@435 | 2087 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2088 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 2089 | |
iveresov@2595 | 2090 | if (entry != NULL) { |
iveresov@2595 | 2091 | *entry = __ pc(); |
iveresov@2595 | 2092 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 2093 | BLOCK_COMMENT("Entry:"); |
duke@435 | 2094 | } |
iveresov@2595 | 2095 | |
iveresov@2595 | 2096 | array_overlap_test(nooverlap_target, Address::times_8); |
duke@435 | 2097 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 2098 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 2099 | // 'from', 'to' and 'qword_count' are now valid |
duke@435 | 2100 | if (is_oop) { |
duke@435 | 2101 | // Save to and count for store barrier |
never@739 | 2102 | __ movptr(saved_count, qword_count); |
duke@435 | 2103 | // No registers are destroyed by this call |
iveresov@2606 | 2104 | gen_write_ref_array_pre_barrier(to, saved_count, dest_uninitialized); |
duke@435 | 2105 | } |
duke@435 | 2106 | |
duke@435 | 2107 | __ jmp(L_copy_32_bytes); |
duke@435 | 2108 | |
duke@435 | 2109 | // Copy trailing qwords |
duke@435 | 2110 | __ BIND(L_copy_8_bytes); |
duke@435 | 2111 | __ movq(rax, Address(from, qword_count, Address::times_8, -8)); |
duke@435 | 2112 | __ movq(Address(to, qword_count, Address::times_8, -8), rax); |
never@739 | 2113 | __ decrement(qword_count); |
duke@435 | 2114 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 2115 | |
duke@435 | 2116 | if (is_oop) { |
duke@435 | 2117 | __ jmp(L_exit); |
duke@435 | 2118 | } else { |
duke@435 | 2119 | restore_arg_regs(); |
never@3314 | 2120 | inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 2121 | __ xorptr(rax, rax); // return 0 |
duke@435 | 2122 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2123 | __ ret(0); |
duke@435 | 2124 | } |
duke@435 | 2125 | |
duke@435 | 2126 | // Copy in 32-bytes chunks |
duke@435 | 2127 | copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes); |
duke@435 | 2128 | |
duke@435 | 2129 | if (is_oop) { |
duke@435 | 2130 | __ BIND(L_exit); |
never@739 | 2131 | __ lea(rcx, Address(to, saved_count, Address::times_8, -8)); |
duke@435 | 2132 | gen_write_ref_array_post_barrier(to, rcx, rax); |
duke@435 | 2133 | } |
duke@435 | 2134 | restore_arg_regs(); |
never@3314 | 2135 | if (is_oop) { |
never@3314 | 2136 | inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free |
never@3314 | 2137 | } else { |
never@3314 | 2138 | inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free |
never@3314 | 2139 | } |
never@739 | 2140 | __ xorptr(rax, rax); // return 0 |
duke@435 | 2141 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2142 | __ ret(0); |
duke@435 | 2143 | |
duke@435 | 2144 | return start; |
duke@435 | 2145 | } |
duke@435 | 2146 | |
duke@435 | 2147 | |
duke@435 | 2148 | // Helper for generating a dynamic type check. |
duke@435 | 2149 | // Smashes no registers. |
duke@435 | 2150 | void generate_type_check(Register sub_klass, |
duke@435 | 2151 | Register super_check_offset, |
duke@435 | 2152 | Register super_klass, |
duke@435 | 2153 | Label& L_success) { |
duke@435 | 2154 | assert_different_registers(sub_klass, super_check_offset, super_klass); |
duke@435 | 2155 | |
duke@435 | 2156 | BLOCK_COMMENT("type_check:"); |
duke@435 | 2157 | |
duke@435 | 2158 | Label L_miss; |
duke@435 | 2159 | |
jrose@1079 | 2160 | __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL, |
jrose@1079 | 2161 | super_check_offset); |
jrose@1079 | 2162 | __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL); |
duke@435 | 2163 | |
duke@435 | 2164 | // Fall through on failure! |
duke@435 | 2165 | __ BIND(L_miss); |
duke@435 | 2166 | } |
duke@435 | 2167 | |
duke@435 | 2168 | // |
duke@435 | 2169 | // Generate checkcasting array copy stub |
duke@435 | 2170 | // |
duke@435 | 2171 | // Input: |
duke@435 | 2172 | // c_rarg0 - source array address |
duke@435 | 2173 | // c_rarg1 - destination array address |
duke@435 | 2174 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 2175 | // c_rarg3 - size_t ckoff (super_check_offset) |
duke@435 | 2176 | // not Win64 |
duke@435 | 2177 | // c_rarg4 - oop ckval (super_klass) |
duke@435 | 2178 | // Win64 |
duke@435 | 2179 | // rsp+40 - oop ckval (super_klass) |
duke@435 | 2180 | // |
duke@435 | 2181 | // Output: |
duke@435 | 2182 | // rax == 0 - success |
duke@435 | 2183 | // rax == -1^K - failure, where K is partial transfer count |
duke@435 | 2184 | // |
iveresov@2606 | 2185 | address generate_checkcast_copy(const char *name, address *entry, |
iveresov@2606 | 2186 | bool dest_uninitialized = false) { |
duke@435 | 2187 | |
duke@435 | 2188 | Label L_load_element, L_store_element, L_do_card_marks, L_done; |
duke@435 | 2189 | |
duke@435 | 2190 | // Input registers (after setup_arg_regs) |
duke@435 | 2191 | const Register from = rdi; // source array address |
duke@435 | 2192 | const Register to = rsi; // destination array address |
duke@435 | 2193 | const Register length = rdx; // elements count |
duke@435 | 2194 | const Register ckoff = rcx; // super_check_offset |
duke@435 | 2195 | const Register ckval = r8; // super_klass |
duke@435 | 2196 | |
duke@435 | 2197 | // Registers used as temps (r13, r14 are save-on-entry) |
duke@435 | 2198 | const Register end_from = from; // source array end address |
duke@435 | 2199 | const Register end_to = r13; // destination array end address |
duke@435 | 2200 | const Register count = rdx; // -(count_remaining) |
duke@435 | 2201 | const Register r14_length = r14; // saved copy of length |
duke@435 | 2202 | // End pointers are inclusive, and if length is not zero they point |
duke@435 | 2203 | // to the last unit copied: end_to[0] := end_from[0] |
duke@435 | 2204 | |
duke@435 | 2205 | const Register rax_oop = rax; // actual oop copied |
duke@435 | 2206 | const Register r11_klass = r11; // oop._klass |
duke@435 | 2207 | |
duke@435 | 2208 | //--------------------------------------------------------------- |
duke@435 | 2209 | // Assembler stub will be used for this call to arraycopy |
duke@435 | 2210 | // if the two arrays are subtypes of Object[] but the |
duke@435 | 2211 | // destination array type is not equal to or a supertype |
duke@435 | 2212 | // of the source type. Each element must be separately |
duke@435 | 2213 | // checked. |
duke@435 | 2214 | |
duke@435 | 2215 | __ align(CodeEntryAlignment); |
duke@435 | 2216 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2217 | address start = __ pc(); |
duke@435 | 2218 | |
duke@435 | 2219 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2220 | |
duke@435 | 2221 | #ifdef ASSERT |
duke@435 | 2222 | // caller guarantees that the arrays really are different |
duke@435 | 2223 | // otherwise, we would have to make conjoint checks |
duke@435 | 2224 | { Label L; |
coleenp@548 | 2225 | array_overlap_test(L, TIMES_OOP); |
duke@435 | 2226 | __ stop("checkcast_copy within a single array"); |
duke@435 | 2227 | __ bind(L); |
duke@435 | 2228 | } |
duke@435 | 2229 | #endif //ASSERT |
duke@435 | 2230 | |
duke@435 | 2231 | setup_arg_regs(4); // from => rdi, to => rsi, length => rdx |
duke@435 | 2232 | // ckoff => rcx, ckval => r8 |
duke@435 | 2233 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 2234 | #ifdef _WIN64 |
duke@435 | 2235 | // last argument (#4) is on stack on Win64 |
twisti@2348 | 2236 | __ movptr(ckval, Address(rsp, 6 * wordSize)); |
duke@435 | 2237 | #endif |
duke@435 | 2238 | |
twisti@2348 | 2239 | // Caller of this entry point must set up the argument registers. |
iveresov@2595 | 2240 | if (entry != NULL) { |
iveresov@2595 | 2241 | *entry = __ pc(); |
iveresov@2595 | 2242 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 2243 | } |
twisti@2348 | 2244 | |
twisti@2348 | 2245 | // allocate spill slots for r13, r14 |
twisti@2348 | 2246 | enum { |
twisti@2348 | 2247 | saved_r13_offset, |
twisti@2348 | 2248 | saved_r14_offset, |
twisti@2348 | 2249 | saved_rbp_offset |
twisti@2348 | 2250 | }; |
twisti@2348 | 2251 | __ subptr(rsp, saved_rbp_offset * wordSize); |
twisti@2348 | 2252 | __ movptr(Address(rsp, saved_r13_offset * wordSize), r13); |
twisti@2348 | 2253 | __ movptr(Address(rsp, saved_r14_offset * wordSize), r14); |
twisti@2348 | 2254 | |
duke@435 | 2255 | // check that int operands are properly extended to size_t |
duke@435 | 2256 | assert_clean_int(length, rax); |
duke@435 | 2257 | assert_clean_int(ckoff, rax); |
duke@435 | 2258 | |
duke@435 | 2259 | #ifdef ASSERT |
duke@435 | 2260 | BLOCK_COMMENT("assert consistent ckoff/ckval"); |
duke@435 | 2261 | // The ckoff and ckval must be mutually consistent, |
duke@435 | 2262 | // even though caller generates both. |
duke@435 | 2263 | { Label L; |
duke@435 | 2264 | int sco_offset = (klassOopDesc::header_size() * HeapWordSize + |
duke@435 | 2265 | Klass::super_check_offset_offset_in_bytes()); |
duke@435 | 2266 | __ cmpl(ckoff, Address(ckval, sco_offset)); |
duke@435 | 2267 | __ jcc(Assembler::equal, L); |
duke@435 | 2268 | __ stop("super_check_offset inconsistent"); |
duke@435 | 2269 | __ bind(L); |
duke@435 | 2270 | } |
duke@435 | 2271 | #endif //ASSERT |
duke@435 | 2272 | |
duke@435 | 2273 | // Loop-invariant addresses. They are exclusive end pointers. |
coleenp@548 | 2274 | Address end_from_addr(from, length, TIMES_OOP, 0); |
coleenp@548 | 2275 | Address end_to_addr(to, length, TIMES_OOP, 0); |
duke@435 | 2276 | // Loop-variant addresses. They assume post-incremented count < 0. |
coleenp@548 | 2277 | Address from_element_addr(end_from, count, TIMES_OOP, 0); |
coleenp@548 | 2278 | Address to_element_addr(end_to, count, TIMES_OOP, 0); |
duke@435 | 2279 | |
iveresov@2606 | 2280 | gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); |
duke@435 | 2281 | |
duke@435 | 2282 | // Copy from low to high addresses, indexed from the end of each array. |
never@739 | 2283 | __ lea(end_from, end_from_addr); |
never@739 | 2284 | __ lea(end_to, end_to_addr); |
never@739 | 2285 | __ movptr(r14_length, length); // save a copy of the length |
never@739 | 2286 | assert(length == count, ""); // else fix next line: |
never@739 | 2287 | __ negptr(count); // negate and test the length |
duke@435 | 2288 | __ jcc(Assembler::notZero, L_load_element); |
duke@435 | 2289 | |
duke@435 | 2290 | // Empty array: Nothing to do. |
never@739 | 2291 | __ xorptr(rax, rax); // return 0 on (trivial) success |
duke@435 | 2292 | __ jmp(L_done); |
duke@435 | 2293 | |
duke@435 | 2294 | // ======== begin loop ======== |
duke@435 | 2295 | // (Loop is rotated; its entry is L_load_element.) |
duke@435 | 2296 | // Loop control: |
duke@435 | 2297 | // for (count = -count; count != 0; count++) |
duke@435 | 2298 | // Base pointers src, dst are biased by 8*(count-1),to last element. |
kvn@1800 | 2299 | __ align(OptoLoopAlignment); |
duke@435 | 2300 | |
duke@435 | 2301 | __ BIND(L_store_element); |
coleenp@548 | 2302 | __ store_heap_oop(to_element_addr, rax_oop); // store the oop |
never@739 | 2303 | __ increment(count); // increment the count toward zero |
duke@435 | 2304 | __ jcc(Assembler::zero, L_do_card_marks); |
duke@435 | 2305 | |
duke@435 | 2306 | // ======== loop entry is here ======== |
duke@435 | 2307 | __ BIND(L_load_element); |
coleenp@548 | 2308 | __ load_heap_oop(rax_oop, from_element_addr); // load the oop |
never@739 | 2309 | __ testptr(rax_oop, rax_oop); |
duke@435 | 2310 | __ jcc(Assembler::zero, L_store_element); |
duke@435 | 2311 | |
coleenp@548 | 2312 | __ load_klass(r11_klass, rax_oop);// query the object klass |
duke@435 | 2313 | generate_type_check(r11_klass, ckoff, ckval, L_store_element); |
duke@435 | 2314 | // ======== end loop ======== |
duke@435 | 2315 | |
duke@435 | 2316 | // It was a real error; we must depend on the caller to finish the job. |
duke@435 | 2317 | // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops. |
duke@435 | 2318 | // Emit GC store barriers for the oops we have copied (r14 + rdx), |
duke@435 | 2319 | // and report their number to the caller. |
duke@435 | 2320 | assert_different_registers(rax, r14_length, count, to, end_to, rcx); |
never@739 | 2321 | __ lea(end_to, to_element_addr); |
ysr@1280 | 2322 | __ addptr(end_to, -heapOopSize); // make an inclusive end pointer |
apetrusenko@797 | 2323 | gen_write_ref_array_post_barrier(to, end_to, rscratch1); |
never@739 | 2324 | __ movptr(rax, r14_length); // original oops |
never@739 | 2325 | __ addptr(rax, count); // K = (original - remaining) oops |
never@739 | 2326 | __ notptr(rax); // report (-1^K) to caller |
duke@435 | 2327 | __ jmp(L_done); |
duke@435 | 2328 | |
duke@435 | 2329 | // Come here on success only. |
duke@435 | 2330 | __ BIND(L_do_card_marks); |
ysr@1280 | 2331 | __ addptr(end_to, -heapOopSize); // make an inclusive end pointer |
apetrusenko@797 | 2332 | gen_write_ref_array_post_barrier(to, end_to, rscratch1); |
never@739 | 2333 | __ xorptr(rax, rax); // return 0 on success |
duke@435 | 2334 | |
duke@435 | 2335 | // Common exit point (success or failure). |
duke@435 | 2336 | __ BIND(L_done); |
never@739 | 2337 | __ movptr(r13, Address(rsp, saved_r13_offset * wordSize)); |
never@739 | 2338 | __ movptr(r14, Address(rsp, saved_r14_offset * wordSize)); |
duke@435 | 2339 | restore_arg_regs(); |
never@3314 | 2340 | inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free |
duke@435 | 2341 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2342 | __ ret(0); |
duke@435 | 2343 | |
duke@435 | 2344 | return start; |
duke@435 | 2345 | } |
duke@435 | 2346 | |
duke@435 | 2347 | // |
duke@435 | 2348 | // Generate 'unsafe' array copy stub |
duke@435 | 2349 | // Though just as safe as the other stubs, it takes an unscaled |
duke@435 | 2350 | // size_t argument instead of an element count. |
duke@435 | 2351 | // |
duke@435 | 2352 | // Input: |
duke@435 | 2353 | // c_rarg0 - source array address |
duke@435 | 2354 | // c_rarg1 - destination array address |
duke@435 | 2355 | // c_rarg2 - byte count, treated as ssize_t, can be zero |
duke@435 | 2356 | // |
duke@435 | 2357 | // Examines the alignment of the operands and dispatches |
duke@435 | 2358 | // to a long, int, short, or byte copy loop. |
duke@435 | 2359 | // |
iveresov@2595 | 2360 | address generate_unsafe_copy(const char *name, |
iveresov@2595 | 2361 | address byte_copy_entry, address short_copy_entry, |
iveresov@2595 | 2362 | address int_copy_entry, address long_copy_entry) { |
duke@435 | 2363 | |
duke@435 | 2364 | Label L_long_aligned, L_int_aligned, L_short_aligned; |
duke@435 | 2365 | |
duke@435 | 2366 | // Input registers (before setup_arg_regs) |
duke@435 | 2367 | const Register from = c_rarg0; // source array address |
duke@435 | 2368 | const Register to = c_rarg1; // destination array address |
duke@435 | 2369 | const Register size = c_rarg2; // byte count (size_t) |
duke@435 | 2370 | |
duke@435 | 2371 | // Register used as a temp |
duke@435 | 2372 | const Register bits = rax; // test copy of low bits |
duke@435 | 2373 | |
duke@435 | 2374 | __ align(CodeEntryAlignment); |
duke@435 | 2375 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2376 | address start = __ pc(); |
duke@435 | 2377 | |
duke@435 | 2378 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2379 | |
duke@435 | 2380 | // bump this on entry, not on exit: |
duke@435 | 2381 | inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr); |
duke@435 | 2382 | |
never@739 | 2383 | __ mov(bits, from); |
never@739 | 2384 | __ orptr(bits, to); |
never@739 | 2385 | __ orptr(bits, size); |
duke@435 | 2386 | |
duke@435 | 2387 | __ testb(bits, BytesPerLong-1); |
duke@435 | 2388 | __ jccb(Assembler::zero, L_long_aligned); |
duke@435 | 2389 | |
duke@435 | 2390 | __ testb(bits, BytesPerInt-1); |
duke@435 | 2391 | __ jccb(Assembler::zero, L_int_aligned); |
duke@435 | 2392 | |
duke@435 | 2393 | __ testb(bits, BytesPerShort-1); |
duke@435 | 2394 | __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry)); |
duke@435 | 2395 | |
duke@435 | 2396 | __ BIND(L_short_aligned); |
never@739 | 2397 | __ shrptr(size, LogBytesPerShort); // size => short_count |
duke@435 | 2398 | __ jump(RuntimeAddress(short_copy_entry)); |
duke@435 | 2399 | |
duke@435 | 2400 | __ BIND(L_int_aligned); |
never@739 | 2401 | __ shrptr(size, LogBytesPerInt); // size => int_count |
duke@435 | 2402 | __ jump(RuntimeAddress(int_copy_entry)); |
duke@435 | 2403 | |
duke@435 | 2404 | __ BIND(L_long_aligned); |
never@739 | 2405 | __ shrptr(size, LogBytesPerLong); // size => qword_count |
duke@435 | 2406 | __ jump(RuntimeAddress(long_copy_entry)); |
duke@435 | 2407 | |
duke@435 | 2408 | return start; |
duke@435 | 2409 | } |
duke@435 | 2410 | |
duke@435 | 2411 | // Perform range checks on the proposed arraycopy. |
duke@435 | 2412 | // Kills temp, but nothing else. |
duke@435 | 2413 | // Also, clean the sign bits of src_pos and dst_pos. |
duke@435 | 2414 | void arraycopy_range_checks(Register src, // source array oop (c_rarg0) |
duke@435 | 2415 | Register src_pos, // source position (c_rarg1) |
duke@435 | 2416 | Register dst, // destination array oo (c_rarg2) |
duke@435 | 2417 | Register dst_pos, // destination position (c_rarg3) |
duke@435 | 2418 | Register length, |
duke@435 | 2419 | Register temp, |
duke@435 | 2420 | Label& L_failed) { |
duke@435 | 2421 | BLOCK_COMMENT("arraycopy_range_checks:"); |
duke@435 | 2422 | |
duke@435 | 2423 | // if (src_pos + length > arrayOop(src)->length()) FAIL; |
duke@435 | 2424 | __ movl(temp, length); |
duke@435 | 2425 | __ addl(temp, src_pos); // src_pos + length |
duke@435 | 2426 | __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes())); |
duke@435 | 2427 | __ jcc(Assembler::above, L_failed); |
duke@435 | 2428 | |
duke@435 | 2429 | // if (dst_pos + length > arrayOop(dst)->length()) FAIL; |
duke@435 | 2430 | __ movl(temp, length); |
duke@435 | 2431 | __ addl(temp, dst_pos); // dst_pos + length |
duke@435 | 2432 | __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes())); |
duke@435 | 2433 | __ jcc(Assembler::above, L_failed); |
duke@435 | 2434 | |
duke@435 | 2435 | // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'. |
duke@435 | 2436 | // Move with sign extension can be used since they are positive. |
duke@435 | 2437 | __ movslq(src_pos, src_pos); |
duke@435 | 2438 | __ movslq(dst_pos, dst_pos); |
duke@435 | 2439 | |
duke@435 | 2440 | BLOCK_COMMENT("arraycopy_range_checks done"); |
duke@435 | 2441 | } |
duke@435 | 2442 | |
duke@435 | 2443 | // |
duke@435 | 2444 | // Generate generic array copy stubs |
duke@435 | 2445 | // |
duke@435 | 2446 | // Input: |
duke@435 | 2447 | // c_rarg0 - src oop |
duke@435 | 2448 | // c_rarg1 - src_pos (32-bits) |
duke@435 | 2449 | // c_rarg2 - dst oop |
duke@435 | 2450 | // c_rarg3 - dst_pos (32-bits) |
duke@435 | 2451 | // not Win64 |
duke@435 | 2452 | // c_rarg4 - element count (32-bits) |
duke@435 | 2453 | // Win64 |
duke@435 | 2454 | // rsp+40 - element count (32-bits) |
duke@435 | 2455 | // |
duke@435 | 2456 | // Output: |
duke@435 | 2457 | // rax == 0 - success |
duke@435 | 2458 | // rax == -1^K - failure, where K is partial transfer count |
duke@435 | 2459 | // |
iveresov@2595 | 2460 | address generate_generic_copy(const char *name, |
iveresov@2595 | 2461 | address byte_copy_entry, address short_copy_entry, |
iveresov@2691 | 2462 | address int_copy_entry, address oop_copy_entry, |
iveresov@2691 | 2463 | address long_copy_entry, address checkcast_copy_entry) { |
duke@435 | 2464 | |
duke@435 | 2465 | Label L_failed, L_failed_0, L_objArray; |
duke@435 | 2466 | Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs; |
duke@435 | 2467 | |
duke@435 | 2468 | // Input registers |
duke@435 | 2469 | const Register src = c_rarg0; // source array oop |
duke@435 | 2470 | const Register src_pos = c_rarg1; // source position |
duke@435 | 2471 | const Register dst = c_rarg2; // destination array oop |
duke@435 | 2472 | const Register dst_pos = c_rarg3; // destination position |
twisti@2348 | 2473 | #ifndef _WIN64 |
twisti@2348 | 2474 | const Register length = c_rarg4; |
duke@435 | 2475 | #else |
twisti@2348 | 2476 | const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64 |
duke@435 | 2477 | #endif |
duke@435 | 2478 | |
duke@435 | 2479 | { int modulus = CodeEntryAlignment; |
duke@435 | 2480 | int target = modulus - 5; // 5 = sizeof jmp(L_failed) |
duke@435 | 2481 | int advance = target - (__ offset() % modulus); |
duke@435 | 2482 | if (advance < 0) advance += modulus; |
duke@435 | 2483 | if (advance > 0) __ nop(advance); |
duke@435 | 2484 | } |
duke@435 | 2485 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2486 | |
duke@435 | 2487 | // Short-hop target to L_failed. Makes for denser prologue code. |
duke@435 | 2488 | __ BIND(L_failed_0); |
duke@435 | 2489 | __ jmp(L_failed); |
duke@435 | 2490 | assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed"); |
duke@435 | 2491 | |
duke@435 | 2492 | __ align(CodeEntryAlignment); |
duke@435 | 2493 | address start = __ pc(); |
duke@435 | 2494 | |
duke@435 | 2495 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2496 | |
duke@435 | 2497 | // bump this on entry, not on exit: |
duke@435 | 2498 | inc_counter_np(SharedRuntime::_generic_array_copy_ctr); |
duke@435 | 2499 | |
duke@435 | 2500 | //----------------------------------------------------------------------- |
duke@435 | 2501 | // Assembler stub will be used for this call to arraycopy |
duke@435 | 2502 | // if the following conditions are met: |
duke@435 | 2503 | // |
duke@435 | 2504 | // (1) src and dst must not be null. |
duke@435 | 2505 | // (2) src_pos must not be negative. |
duke@435 | 2506 | // (3) dst_pos must not be negative. |
duke@435 | 2507 | // (4) length must not be negative. |
duke@435 | 2508 | // (5) src klass and dst klass should be the same and not NULL. |
duke@435 | 2509 | // (6) src and dst should be arrays. |
duke@435 | 2510 | // (7) src_pos + length must not exceed length of src. |
duke@435 | 2511 | // (8) dst_pos + length must not exceed length of dst. |
duke@435 | 2512 | // |
duke@435 | 2513 | |
duke@435 | 2514 | // if (src == NULL) return -1; |
never@739 | 2515 | __ testptr(src, src); // src oop |
duke@435 | 2516 | size_t j1off = __ offset(); |
duke@435 | 2517 | __ jccb(Assembler::zero, L_failed_0); |
duke@435 | 2518 | |
duke@435 | 2519 | // if (src_pos < 0) return -1; |
duke@435 | 2520 | __ testl(src_pos, src_pos); // src_pos (32-bits) |
duke@435 | 2521 | __ jccb(Assembler::negative, L_failed_0); |
duke@435 | 2522 | |
duke@435 | 2523 | // if (dst == NULL) return -1; |
never@739 | 2524 | __ testptr(dst, dst); // dst oop |
duke@435 | 2525 | __ jccb(Assembler::zero, L_failed_0); |
duke@435 | 2526 | |
duke@435 | 2527 | // if (dst_pos < 0) return -1; |
duke@435 | 2528 | __ testl(dst_pos, dst_pos); // dst_pos (32-bits) |
duke@435 | 2529 | size_t j4off = __ offset(); |
duke@435 | 2530 | __ jccb(Assembler::negative, L_failed_0); |
duke@435 | 2531 | |
duke@435 | 2532 | // The first four tests are very dense code, |
duke@435 | 2533 | // but not quite dense enough to put four |
duke@435 | 2534 | // jumps in a 16-byte instruction fetch buffer. |
duke@435 | 2535 | // That's good, because some branch predicters |
duke@435 | 2536 | // do not like jumps so close together. |
duke@435 | 2537 | // Make sure of this. |
duke@435 | 2538 | guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps"); |
duke@435 | 2539 | |
duke@435 | 2540 | // registers used as temp |
duke@435 | 2541 | const Register r11_length = r11; // elements count to copy |
duke@435 | 2542 | const Register r10_src_klass = r10; // array klass |
duke@435 | 2543 | |
duke@435 | 2544 | // if (length < 0) return -1; |
twisti@2348 | 2545 | __ movl(r11_length, length); // length (elements count, 32-bits value) |
duke@435 | 2546 | __ testl(r11_length, r11_length); |
duke@435 | 2547 | __ jccb(Assembler::negative, L_failed_0); |
duke@435 | 2548 | |
coleenp@548 | 2549 | __ load_klass(r10_src_klass, src); |
duke@435 | 2550 | #ifdef ASSERT |
duke@435 | 2551 | // assert(src->klass() != NULL); |
twisti@2348 | 2552 | { |
twisti@2348 | 2553 | BLOCK_COMMENT("assert klasses not null {"); |
twisti@2348 | 2554 | Label L1, L2; |
never@739 | 2555 | __ testptr(r10_src_klass, r10_src_klass); |
duke@435 | 2556 | __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL |
duke@435 | 2557 | __ bind(L1); |
duke@435 | 2558 | __ stop("broken null klass"); |
duke@435 | 2559 | __ bind(L2); |
twisti@2348 | 2560 | __ load_klass(rax, dst); |
twisti@2348 | 2561 | __ cmpq(rax, 0); |
duke@435 | 2562 | __ jcc(Assembler::equal, L1); // this would be broken also |
twisti@2348 | 2563 | BLOCK_COMMENT("} assert klasses not null done"); |
duke@435 | 2564 | } |
duke@435 | 2565 | #endif |
duke@435 | 2566 | |
duke@435 | 2567 | // Load layout helper (32-bits) |
duke@435 | 2568 | // |
duke@435 | 2569 | // |array_tag| | header_size | element_type | |log2_element_size| |
duke@435 | 2570 | // 32 30 24 16 8 2 0 |
duke@435 | 2571 | // |
duke@435 | 2572 | // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 |
duke@435 | 2573 | // |
duke@435 | 2574 | |
twisti@2348 | 2575 | const int lh_offset = klassOopDesc::header_size() * HeapWordSize + |
twisti@2348 | 2576 | Klass::layout_helper_offset_in_bytes(); |
twisti@2348 | 2577 | |
twisti@2348 | 2578 | // Handle objArrays completely differently... |
twisti@2348 | 2579 | const jint objArray_lh = Klass::array_layout_helper(T_OBJECT); |
twisti@2348 | 2580 | __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh); |
twisti@2348 | 2581 | __ jcc(Assembler::equal, L_objArray); |
twisti@2348 | 2582 | |
twisti@2348 | 2583 | // if (src->klass() != dst->klass()) return -1; |
twisti@2348 | 2584 | __ load_klass(rax, dst); |
twisti@2348 | 2585 | __ cmpq(r10_src_klass, rax); |
twisti@2348 | 2586 | __ jcc(Assembler::notEqual, L_failed); |
duke@435 | 2587 | |
duke@435 | 2588 | const Register rax_lh = rax; // layout helper |
duke@435 | 2589 | __ movl(rax_lh, Address(r10_src_klass, lh_offset)); |
duke@435 | 2590 | |
duke@435 | 2591 | // if (!src->is_Array()) return -1; |
duke@435 | 2592 | __ cmpl(rax_lh, Klass::_lh_neutral_value); |
duke@435 | 2593 | __ jcc(Assembler::greaterEqual, L_failed); |
duke@435 | 2594 | |
duke@435 | 2595 | // At this point, it is known to be a typeArray (array_tag 0x3). |
duke@435 | 2596 | #ifdef ASSERT |
twisti@2348 | 2597 | { |
twisti@2348 | 2598 | BLOCK_COMMENT("assert primitive array {"); |
twisti@2348 | 2599 | Label L; |
duke@435 | 2600 | __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift)); |
duke@435 | 2601 | __ jcc(Assembler::greaterEqual, L); |
duke@435 | 2602 | __ stop("must be a primitive array"); |
duke@435 | 2603 | __ bind(L); |
twisti@2348 | 2604 | BLOCK_COMMENT("} assert primitive array done"); |
duke@435 | 2605 | } |
duke@435 | 2606 | #endif |
duke@435 | 2607 | |
duke@435 | 2608 | arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, |
duke@435 | 2609 | r10, L_failed); |
duke@435 | 2610 | |
duke@435 | 2611 | // typeArrayKlass |
duke@435 | 2612 | // |
duke@435 | 2613 | // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); |
duke@435 | 2614 | // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); |
duke@435 | 2615 | // |
duke@435 | 2616 | |
duke@435 | 2617 | const Register r10_offset = r10; // array offset |
duke@435 | 2618 | const Register rax_elsize = rax_lh; // element size |
duke@435 | 2619 | |
duke@435 | 2620 | __ movl(r10_offset, rax_lh); |
duke@435 | 2621 | __ shrl(r10_offset, Klass::_lh_header_size_shift); |
never@739 | 2622 | __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset |
never@739 | 2623 | __ addptr(src, r10_offset); // src array offset |
never@739 | 2624 | __ addptr(dst, r10_offset); // dst array offset |
duke@435 | 2625 | BLOCK_COMMENT("choose copy loop based on element size"); |
duke@435 | 2626 | __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize |
duke@435 | 2627 | |
duke@435 | 2628 | // next registers should be set before the jump to corresponding stub |
duke@435 | 2629 | const Register from = c_rarg0; // source array address |
duke@435 | 2630 | const Register to = c_rarg1; // destination array address |
duke@435 | 2631 | const Register count = c_rarg2; // elements count |
duke@435 | 2632 | |
duke@435 | 2633 | // 'from', 'to', 'count' registers should be set in such order |
duke@435 | 2634 | // since they are the same as 'src', 'src_pos', 'dst'. |
duke@435 | 2635 | |
duke@435 | 2636 | __ BIND(L_copy_bytes); |
duke@435 | 2637 | __ cmpl(rax_elsize, 0); |
duke@435 | 2638 | __ jccb(Assembler::notEqual, L_copy_shorts); |
never@739 | 2639 | __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr |
never@739 | 2640 | __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr |
never@739 | 2641 | __ movl2ptr(count, r11_length); // length |
duke@435 | 2642 | __ jump(RuntimeAddress(byte_copy_entry)); |
duke@435 | 2643 | |
duke@435 | 2644 | __ BIND(L_copy_shorts); |
duke@435 | 2645 | __ cmpl(rax_elsize, LogBytesPerShort); |
duke@435 | 2646 | __ jccb(Assembler::notEqual, L_copy_ints); |
never@739 | 2647 | __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr |
never@739 | 2648 | __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr |
never@739 | 2649 | __ movl2ptr(count, r11_length); // length |
duke@435 | 2650 | __ jump(RuntimeAddress(short_copy_entry)); |
duke@435 | 2651 | |
duke@435 | 2652 | __ BIND(L_copy_ints); |
duke@435 | 2653 | __ cmpl(rax_elsize, LogBytesPerInt); |
duke@435 | 2654 | __ jccb(Assembler::notEqual, L_copy_longs); |
never@739 | 2655 | __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr |
never@739 | 2656 | __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr |
never@739 | 2657 | __ movl2ptr(count, r11_length); // length |
duke@435 | 2658 | __ jump(RuntimeAddress(int_copy_entry)); |
duke@435 | 2659 | |
duke@435 | 2660 | __ BIND(L_copy_longs); |
duke@435 | 2661 | #ifdef ASSERT |
twisti@2348 | 2662 | { |
twisti@2348 | 2663 | BLOCK_COMMENT("assert long copy {"); |
twisti@2348 | 2664 | Label L; |
duke@435 | 2665 | __ cmpl(rax_elsize, LogBytesPerLong); |
duke@435 | 2666 | __ jcc(Assembler::equal, L); |
duke@435 | 2667 | __ stop("must be long copy, but elsize is wrong"); |
duke@435 | 2668 | __ bind(L); |
twisti@2348 | 2669 | BLOCK_COMMENT("} assert long copy done"); |
duke@435 | 2670 | } |
duke@435 | 2671 | #endif |
never@739 | 2672 | __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr |
never@739 | 2673 | __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr |
never@739 | 2674 | __ movl2ptr(count, r11_length); // length |
duke@435 | 2675 | __ jump(RuntimeAddress(long_copy_entry)); |
duke@435 | 2676 | |
duke@435 | 2677 | // objArrayKlass |
duke@435 | 2678 | __ BIND(L_objArray); |
twisti@2348 | 2679 | // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos] |
duke@435 | 2680 | |
duke@435 | 2681 | Label L_plain_copy, L_checkcast_copy; |
duke@435 | 2682 | // test array classes for subtyping |
twisti@2348 | 2683 | __ load_klass(rax, dst); |
twisti@2348 | 2684 | __ cmpq(r10_src_klass, rax); // usual case is exact equality |
duke@435 | 2685 | __ jcc(Assembler::notEqual, L_checkcast_copy); |
duke@435 | 2686 | |
duke@435 | 2687 | // Identically typed arrays can be copied without element-wise checks. |
duke@435 | 2688 | arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, |
duke@435 | 2689 | r10, L_failed); |
duke@435 | 2690 | |
never@739 | 2691 | __ lea(from, Address(src, src_pos, TIMES_OOP, |
duke@435 | 2692 | arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr |
never@739 | 2693 | __ lea(to, Address(dst, dst_pos, TIMES_OOP, |
never@739 | 2694 | arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr |
never@739 | 2695 | __ movl2ptr(count, r11_length); // length |
duke@435 | 2696 | __ BIND(L_plain_copy); |
duke@435 | 2697 | __ jump(RuntimeAddress(oop_copy_entry)); |
duke@435 | 2698 | |
duke@435 | 2699 | __ BIND(L_checkcast_copy); |
twisti@2348 | 2700 | // live at this point: r10_src_klass, r11_length, rax (dst_klass) |
duke@435 | 2701 | { |
duke@435 | 2702 | // Before looking at dst.length, make sure dst is also an objArray. |
twisti@2348 | 2703 | __ cmpl(Address(rax, lh_offset), objArray_lh); |
duke@435 | 2704 | __ jcc(Assembler::notEqual, L_failed); |
duke@435 | 2705 | |
duke@435 | 2706 | // It is safe to examine both src.length and dst.length. |
duke@435 | 2707 | arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, |
duke@435 | 2708 | rax, L_failed); |
twisti@2348 | 2709 | |
twisti@2348 | 2710 | const Register r11_dst_klass = r11; |
coleenp@548 | 2711 | __ load_klass(r11_dst_klass, dst); // reload |
duke@435 | 2712 | |
duke@435 | 2713 | // Marshal the base address arguments now, freeing registers. |
never@739 | 2714 | __ lea(from, Address(src, src_pos, TIMES_OOP, |
duke@435 | 2715 | arrayOopDesc::base_offset_in_bytes(T_OBJECT))); |
never@739 | 2716 | __ lea(to, Address(dst, dst_pos, TIMES_OOP, |
duke@435 | 2717 | arrayOopDesc::base_offset_in_bytes(T_OBJECT))); |
twisti@2348 | 2718 | __ movl(count, length); // length (reloaded) |
duke@435 | 2719 | Register sco_temp = c_rarg3; // this register is free now |
duke@435 | 2720 | assert_different_registers(from, to, count, sco_temp, |
duke@435 | 2721 | r11_dst_klass, r10_src_klass); |
duke@435 | 2722 | assert_clean_int(count, sco_temp); |
duke@435 | 2723 | |
duke@435 | 2724 | // Generate the type check. |
twisti@2348 | 2725 | const int sco_offset = (klassOopDesc::header_size() * HeapWordSize + |
twisti@2348 | 2726 | Klass::super_check_offset_offset_in_bytes()); |
duke@435 | 2727 | __ movl(sco_temp, Address(r11_dst_klass, sco_offset)); |
duke@435 | 2728 | assert_clean_int(sco_temp, rax); |
duke@435 | 2729 | generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy); |
duke@435 | 2730 | |
duke@435 | 2731 | // Fetch destination element klass from the objArrayKlass header. |
duke@435 | 2732 | int ek_offset = (klassOopDesc::header_size() * HeapWordSize + |
duke@435 | 2733 | objArrayKlass::element_klass_offset_in_bytes()); |
never@739 | 2734 | __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset)); |
twisti@2348 | 2735 | __ movl( sco_temp, Address(r11_dst_klass, sco_offset)); |
duke@435 | 2736 | assert_clean_int(sco_temp, rax); |
duke@435 | 2737 | |
duke@435 | 2738 | // the checkcast_copy loop needs two extra arguments: |
duke@435 | 2739 | assert(c_rarg3 == sco_temp, "#3 already in place"); |
twisti@2348 | 2740 | // Set up arguments for checkcast_copy_entry. |
twisti@2348 | 2741 | setup_arg_regs(4); |
twisti@2348 | 2742 | __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris |
duke@435 | 2743 | __ jump(RuntimeAddress(checkcast_copy_entry)); |
duke@435 | 2744 | } |
duke@435 | 2745 | |
duke@435 | 2746 | __ BIND(L_failed); |
never@739 | 2747 | __ xorptr(rax, rax); |
never@739 | 2748 | __ notptr(rax); // return -1 |
duke@435 | 2749 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2750 | __ ret(0); |
duke@435 | 2751 | |
duke@435 | 2752 | return start; |
duke@435 | 2753 | } |
duke@435 | 2754 | |
duke@435 | 2755 | void generate_arraycopy_stubs() { |
iveresov@2595 | 2756 | address entry; |
iveresov@2595 | 2757 | address entry_jbyte_arraycopy; |
iveresov@2595 | 2758 | address entry_jshort_arraycopy; |
iveresov@2595 | 2759 | address entry_jint_arraycopy; |
iveresov@2595 | 2760 | address entry_oop_arraycopy; |
iveresov@2595 | 2761 | address entry_jlong_arraycopy; |
iveresov@2595 | 2762 | address entry_checkcast_arraycopy; |
iveresov@2595 | 2763 | |
iveresov@2595 | 2764 | StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry, |
iveresov@2595 | 2765 | "jbyte_disjoint_arraycopy"); |
iveresov@2595 | 2766 | StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy, |
iveresov@2595 | 2767 | "jbyte_arraycopy"); |
iveresov@2595 | 2768 | |
iveresov@2595 | 2769 | StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry, |
iveresov@2595 | 2770 | "jshort_disjoint_arraycopy"); |
iveresov@2595 | 2771 | StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy, |
iveresov@2595 | 2772 | "jshort_arraycopy"); |
iveresov@2595 | 2773 | |
iveresov@2595 | 2774 | StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry, |
iveresov@2595 | 2775 | "jint_disjoint_arraycopy"); |
iveresov@2595 | 2776 | StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry, |
iveresov@2595 | 2777 | &entry_jint_arraycopy, "jint_arraycopy"); |
iveresov@2595 | 2778 | |
iveresov@2595 | 2779 | StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry, |
iveresov@2595 | 2780 | "jlong_disjoint_arraycopy"); |
iveresov@2595 | 2781 | StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry, |
iveresov@2595 | 2782 | &entry_jlong_arraycopy, "jlong_arraycopy"); |
duke@435 | 2783 | |
coleenp@548 | 2784 | |
coleenp@548 | 2785 | if (UseCompressedOops) { |
iveresov@2595 | 2786 | StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry, |
iveresov@2595 | 2787 | "oop_disjoint_arraycopy"); |
iveresov@2595 | 2788 | StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry, |
iveresov@2595 | 2789 | &entry_oop_arraycopy, "oop_arraycopy"); |
iveresov@2606 | 2790 | StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry, |
iveresov@2606 | 2791 | "oop_disjoint_arraycopy_uninit", |
iveresov@2606 | 2792 | /*dest_uninitialized*/true); |
iveresov@2606 | 2793 | StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry, |
iveresov@2606 | 2794 | NULL, "oop_arraycopy_uninit", |
iveresov@2606 | 2795 | /*dest_uninitialized*/true); |
coleenp@548 | 2796 | } else { |
iveresov@2595 | 2797 | StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry, |
iveresov@2595 | 2798 | "oop_disjoint_arraycopy"); |
iveresov@2595 | 2799 | StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry, |
iveresov@2595 | 2800 | &entry_oop_arraycopy, "oop_arraycopy"); |
iveresov@2606 | 2801 | StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry, |
iveresov@2606 | 2802 | "oop_disjoint_arraycopy_uninit", |
iveresov@2606 | 2803 | /*dest_uninitialized*/true); |
iveresov@2606 | 2804 | StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry, |
iveresov@2606 | 2805 | NULL, "oop_arraycopy_uninit", |
iveresov@2606 | 2806 | /*dest_uninitialized*/true); |
coleenp@548 | 2807 | } |
duke@435 | 2808 | |
iveresov@2606 | 2809 | StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy); |
iveresov@2606 | 2810 | StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, |
iveresov@2606 | 2811 | /*dest_uninitialized*/true); |
iveresov@2606 | 2812 | |
iveresov@2595 | 2813 | StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy", |
iveresov@2595 | 2814 | entry_jbyte_arraycopy, |
iveresov@2595 | 2815 | entry_jshort_arraycopy, |
iveresov@2595 | 2816 | entry_jint_arraycopy, |
iveresov@2595 | 2817 | entry_jlong_arraycopy); |
iveresov@2595 | 2818 | StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy", |
iveresov@2595 | 2819 | entry_jbyte_arraycopy, |
iveresov@2595 | 2820 | entry_jshort_arraycopy, |
iveresov@2595 | 2821 | entry_jint_arraycopy, |
iveresov@2595 | 2822 | entry_oop_arraycopy, |
iveresov@2595 | 2823 | entry_jlong_arraycopy, |
iveresov@2595 | 2824 | entry_checkcast_arraycopy); |
duke@435 | 2825 | |
never@2118 | 2826 | StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill"); |
never@2118 | 2827 | StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill"); |
never@2118 | 2828 | StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill"); |
never@2118 | 2829 | StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill"); |
never@2118 | 2830 | StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill"); |
never@2118 | 2831 | StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill"); |
never@2118 | 2832 | |
duke@435 | 2833 | // We don't generate specialized code for HeapWord-aligned source |
duke@435 | 2834 | // arrays, so just use the code we've already generated |
duke@435 | 2835 | StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy; |
duke@435 | 2836 | StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy; |
duke@435 | 2837 | |
duke@435 | 2838 | StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy; |
duke@435 | 2839 | StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy; |
duke@435 | 2840 | |
duke@435 | 2841 | StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy; |
duke@435 | 2842 | StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy; |
duke@435 | 2843 | |
duke@435 | 2844 | StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy; |
duke@435 | 2845 | StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy; |
duke@435 | 2846 | |
duke@435 | 2847 | StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy; |
duke@435 | 2848 | StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; |
iveresov@2606 | 2849 | |
iveresov@2606 | 2850 | StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit; |
iveresov@2606 | 2851 | StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit; |
duke@435 | 2852 | } |
duke@435 | 2853 | |
never@1609 | 2854 | void generate_math_stubs() { |
never@1609 | 2855 | { |
never@1609 | 2856 | StubCodeMark mark(this, "StubRoutines", "log"); |
never@1609 | 2857 | StubRoutines::_intrinsic_log = (double (*)(double)) __ pc(); |
never@1609 | 2858 | |
never@1609 | 2859 | __ subq(rsp, 8); |
never@1609 | 2860 | __ movdbl(Address(rsp, 0), xmm0); |
never@1609 | 2861 | __ fld_d(Address(rsp, 0)); |
never@1609 | 2862 | __ flog(); |
never@1609 | 2863 | __ fstp_d(Address(rsp, 0)); |
never@1609 | 2864 | __ movdbl(xmm0, Address(rsp, 0)); |
never@1609 | 2865 | __ addq(rsp, 8); |
never@1609 | 2866 | __ ret(0); |
never@1609 | 2867 | } |
never@1609 | 2868 | { |
never@1609 | 2869 | StubCodeMark mark(this, "StubRoutines", "log10"); |
never@1609 | 2870 | StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc(); |
never@1609 | 2871 | |
never@1609 | 2872 | __ subq(rsp, 8); |
never@1609 | 2873 | __ movdbl(Address(rsp, 0), xmm0); |
never@1609 | 2874 | __ fld_d(Address(rsp, 0)); |
never@1609 | 2875 | __ flog10(); |
never@1609 | 2876 | __ fstp_d(Address(rsp, 0)); |
never@1609 | 2877 | __ movdbl(xmm0, Address(rsp, 0)); |
never@1609 | 2878 | __ addq(rsp, 8); |
never@1609 | 2879 | __ ret(0); |
never@1609 | 2880 | } |
never@1609 | 2881 | { |
never@1609 | 2882 | StubCodeMark mark(this, "StubRoutines", "sin"); |
never@1609 | 2883 | StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc(); |
never@1609 | 2884 | |
never@1609 | 2885 | __ subq(rsp, 8); |
never@1609 | 2886 | __ movdbl(Address(rsp, 0), xmm0); |
never@1609 | 2887 | __ fld_d(Address(rsp, 0)); |
never@1609 | 2888 | __ trigfunc('s'); |
never@1609 | 2889 | __ fstp_d(Address(rsp, 0)); |
never@1609 | 2890 | __ movdbl(xmm0, Address(rsp, 0)); |
never@1609 | 2891 | __ addq(rsp, 8); |
never@1609 | 2892 | __ ret(0); |
never@1609 | 2893 | } |
never@1609 | 2894 | { |
never@1609 | 2895 | StubCodeMark mark(this, "StubRoutines", "cos"); |
never@1609 | 2896 | StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc(); |
never@1609 | 2897 | |
never@1609 | 2898 | __ subq(rsp, 8); |
never@1609 | 2899 | __ movdbl(Address(rsp, 0), xmm0); |
never@1609 | 2900 | __ fld_d(Address(rsp, 0)); |
never@1609 | 2901 | __ trigfunc('c'); |
never@1609 | 2902 | __ fstp_d(Address(rsp, 0)); |
never@1609 | 2903 | __ movdbl(xmm0, Address(rsp, 0)); |
never@1609 | 2904 | __ addq(rsp, 8); |
never@1609 | 2905 | __ ret(0); |
never@1609 | 2906 | } |
never@1609 | 2907 | { |
never@1609 | 2908 | StubCodeMark mark(this, "StubRoutines", "tan"); |
never@1609 | 2909 | StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc(); |
never@1609 | 2910 | |
never@1609 | 2911 | __ subq(rsp, 8); |
never@1609 | 2912 | __ movdbl(Address(rsp, 0), xmm0); |
never@1609 | 2913 | __ fld_d(Address(rsp, 0)); |
never@1609 | 2914 | __ trigfunc('t'); |
never@1609 | 2915 | __ fstp_d(Address(rsp, 0)); |
never@1609 | 2916 | __ movdbl(xmm0, Address(rsp, 0)); |
never@1609 | 2917 | __ addq(rsp, 8); |
never@1609 | 2918 | __ ret(0); |
never@1609 | 2919 | } |
never@1609 | 2920 | |
never@1609 | 2921 | // The intrinsic version of these seem to return the same value as |
never@1609 | 2922 | // the strict version. |
never@1609 | 2923 | StubRoutines::_intrinsic_exp = SharedRuntime::dexp; |
never@1609 | 2924 | StubRoutines::_intrinsic_pow = SharedRuntime::dpow; |
never@1609 | 2925 | } |
never@1609 | 2926 | |
duke@435 | 2927 | #undef __ |
duke@435 | 2928 | #define __ masm-> |
duke@435 | 2929 | |
duke@435 | 2930 | // Continuation point for throwing of implicit exceptions that are |
duke@435 | 2931 | // not handled in the current activation. Fabricates an exception |
duke@435 | 2932 | // oop and initiates normal exception dispatching in this |
duke@435 | 2933 | // frame. Since we need to preserve callee-saved values (currently |
duke@435 | 2934 | // only for C2, but done for C1 as well) we need a callee-saved oop |
duke@435 | 2935 | // map and therefore have to make these stubs into RuntimeStubs |
duke@435 | 2936 | // rather than BufferBlobs. If the compiler needs all registers to |
duke@435 | 2937 | // be preserved between the fault point and the exception handler |
duke@435 | 2938 | // then it must assume responsibility for that in |
duke@435 | 2939 | // AbstractCompiler::continuation_for_implicit_null_exception or |
duke@435 | 2940 | // continuation_for_implicit_division_by_zero_exception. All other |
duke@435 | 2941 | // implicit exceptions (e.g., NullPointerException or |
duke@435 | 2942 | // AbstractMethodError on entry) are either at call sites or |
duke@435 | 2943 | // otherwise assume that stack unwinding will be initiated, so |
duke@435 | 2944 | // caller saved registers were assumed volatile in the compiler. |
duke@435 | 2945 | address generate_throw_exception(const char* name, |
duke@435 | 2946 | address runtime_entry, |
never@2978 | 2947 | Register arg1 = noreg, |
never@2978 | 2948 | Register arg2 = noreg) { |
duke@435 | 2949 | // Information about frame layout at time of blocking runtime call. |
duke@435 | 2950 | // Note that we only have to preserve callee-saved registers since |
duke@435 | 2951 | // the compilers are responsible for supplying a continuation point |
duke@435 | 2952 | // if they expect all registers to be preserved. |
duke@435 | 2953 | enum layout { |
duke@435 | 2954 | rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt, |
duke@435 | 2955 | rbp_off2, |
duke@435 | 2956 | return_off, |
duke@435 | 2957 | return_off2, |
duke@435 | 2958 | framesize // inclusive of return address |
duke@435 | 2959 | }; |
duke@435 | 2960 | |
duke@435 | 2961 | int insts_size = 512; |
duke@435 | 2962 | int locs_size = 64; |
duke@435 | 2963 | |
duke@435 | 2964 | CodeBuffer code(name, insts_size, locs_size); |
duke@435 | 2965 | OopMapSet* oop_maps = new OopMapSet(); |
duke@435 | 2966 | MacroAssembler* masm = new MacroAssembler(&code); |
duke@435 | 2967 | |
duke@435 | 2968 | address start = __ pc(); |
duke@435 | 2969 | |
duke@435 | 2970 | // This is an inlined and slightly modified version of call_VM |
duke@435 | 2971 | // which has the ability to fetch the return PC out of |
duke@435 | 2972 | // thread-local storage and also sets up last_Java_sp slightly |
duke@435 | 2973 | // differently than the real call_VM |
duke@435 | 2974 | |
duke@435 | 2975 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2976 | |
duke@435 | 2977 | assert(is_even(framesize/2), "sp not 16-byte aligned"); |
duke@435 | 2978 | |
duke@435 | 2979 | // return address and rbp are already in place |
never@739 | 2980 | __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog |
duke@435 | 2981 | |
duke@435 | 2982 | int frame_complete = __ pc() - start; |
duke@435 | 2983 | |
duke@435 | 2984 | // Set up last_Java_sp and last_Java_fp |
duke@435 | 2985 | __ set_last_Java_frame(rsp, rbp, NULL); |
duke@435 | 2986 | |
duke@435 | 2987 | // Call runtime |
never@2978 | 2988 | if (arg1 != noreg) { |
never@2978 | 2989 | assert(arg2 != c_rarg1, "clobbered"); |
never@2978 | 2990 | __ movptr(c_rarg1, arg1); |
never@2978 | 2991 | } |
never@2978 | 2992 | if (arg2 != noreg) { |
never@2978 | 2993 | __ movptr(c_rarg2, arg2); |
never@2978 | 2994 | } |
never@739 | 2995 | __ movptr(c_rarg0, r15_thread); |
duke@435 | 2996 | BLOCK_COMMENT("call runtime_entry"); |
duke@435 | 2997 | __ call(RuntimeAddress(runtime_entry)); |
duke@435 | 2998 | |
duke@435 | 2999 | // Generate oop map |
duke@435 | 3000 | OopMap* map = new OopMap(framesize, 0); |
duke@435 | 3001 | |
duke@435 | 3002 | oop_maps->add_gc_map(__ pc() - start, map); |
duke@435 | 3003 | |
duke@435 | 3004 | __ reset_last_Java_frame(true, false); |
duke@435 | 3005 | |
duke@435 | 3006 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 3007 | |
duke@435 | 3008 | // check for pending exceptions |
duke@435 | 3009 | #ifdef ASSERT |
duke@435 | 3010 | Label L; |
never@739 | 3011 | __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), |
never@739 | 3012 | (int32_t) NULL_WORD); |
duke@435 | 3013 | __ jcc(Assembler::notEqual, L); |
duke@435 | 3014 | __ should_not_reach_here(); |
duke@435 | 3015 | __ bind(L); |
duke@435 | 3016 | #endif // ASSERT |
duke@435 | 3017 | __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); |
duke@435 | 3018 | |
duke@435 | 3019 | |
duke@435 | 3020 | // codeBlob framesize is in words (not VMRegImpl::slot_size) |
duke@435 | 3021 | RuntimeStub* stub = |
duke@435 | 3022 | RuntimeStub::new_runtime_stub(name, |
duke@435 | 3023 | &code, |
duke@435 | 3024 | frame_complete, |
duke@435 | 3025 | (framesize >> (LogBytesPerWord - LogBytesPerInt)), |
duke@435 | 3026 | oop_maps, false); |
duke@435 | 3027 | return stub->entry_point(); |
duke@435 | 3028 | } |
duke@435 | 3029 | |
duke@435 | 3030 | // Initialization |
duke@435 | 3031 | void generate_initial() { |
duke@435 | 3032 | // Generates all stubs and initializes the entry points |
duke@435 | 3033 | |
duke@435 | 3034 | // This platform-specific stub is needed by generate_call_stub() |
never@739 | 3035 | StubRoutines::x86::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80); |
duke@435 | 3036 | |
duke@435 | 3037 | // entry points that exist in all platforms Note: This is code |
duke@435 | 3038 | // that could be shared among different platforms - however the |
duke@435 | 3039 | // benefit seems to be smaller than the disadvantage of having a |
duke@435 | 3040 | // much more complicated generator structure. See also comment in |
duke@435 | 3041 | // stubRoutines.hpp. |
duke@435 | 3042 | |
duke@435 | 3043 | StubRoutines::_forward_exception_entry = generate_forward_exception(); |
duke@435 | 3044 | |
duke@435 | 3045 | StubRoutines::_call_stub_entry = |
duke@435 | 3046 | generate_call_stub(StubRoutines::_call_stub_return_address); |
duke@435 | 3047 | |
duke@435 | 3048 | // is referenced by megamorphic call |
duke@435 | 3049 | StubRoutines::_catch_exception_entry = generate_catch_exception(); |
duke@435 | 3050 | |
duke@435 | 3051 | // atomic calls |
duke@435 | 3052 | StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); |
duke@435 | 3053 | StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr(); |
duke@435 | 3054 | StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg(); |
duke@435 | 3055 | StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long(); |
duke@435 | 3056 | StubRoutines::_atomic_add_entry = generate_atomic_add(); |
duke@435 | 3057 | StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr(); |
duke@435 | 3058 | StubRoutines::_fence_entry = generate_orderaccess_fence(); |
duke@435 | 3059 | |
duke@435 | 3060 | StubRoutines::_handler_for_unsafe_access_entry = |
duke@435 | 3061 | generate_handler_for_unsafe_access(); |
duke@435 | 3062 | |
duke@435 | 3063 | // platform dependent |
never@739 | 3064 | StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp(); |
never@739 | 3065 | |
never@739 | 3066 | StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr(); |
never@2978 | 3067 | |
never@2978 | 3068 | // Build this early so it's available for the interpreter. Stub |
never@2978 | 3069 | // expects the required and actual types as register arguments in |
never@2978 | 3070 | // j_rarg0 and j_rarg1 respectively. |
never@2978 | 3071 | StubRoutines::_throw_WrongMethodTypeException_entry = |
never@2978 | 3072 | generate_throw_exception("WrongMethodTypeException throw_exception", |
never@2978 | 3073 | CAST_FROM_FN_PTR(address, SharedRuntime::throw_WrongMethodTypeException), |
never@3136 | 3074 | rax, rcx); |
duke@435 | 3075 | } |
duke@435 | 3076 | |
duke@435 | 3077 | void generate_all() { |
duke@435 | 3078 | // Generates all stubs and initializes the entry points |
duke@435 | 3079 | |
duke@435 | 3080 | // These entry points require SharedInfo::stack0 to be set up in |
duke@435 | 3081 | // non-core builds and need to be relocatable, so they each |
duke@435 | 3082 | // fabricate a RuntimeStub internally. |
duke@435 | 3083 | StubRoutines::_throw_AbstractMethodError_entry = |
duke@435 | 3084 | generate_throw_exception("AbstractMethodError throw_exception", |
duke@435 | 3085 | CAST_FROM_FN_PTR(address, |
duke@435 | 3086 | SharedRuntime:: |
never@3136 | 3087 | throw_AbstractMethodError)); |
duke@435 | 3088 | |
dcubed@451 | 3089 | StubRoutines::_throw_IncompatibleClassChangeError_entry = |
dcubed@451 | 3090 | generate_throw_exception("IncompatibleClassChangeError throw_exception", |
dcubed@451 | 3091 | CAST_FROM_FN_PTR(address, |
dcubed@451 | 3092 | SharedRuntime:: |
never@3136 | 3093 | throw_IncompatibleClassChangeError)); |
duke@435 | 3094 | |
duke@435 | 3095 | StubRoutines::_throw_NullPointerException_at_call_entry = |
duke@435 | 3096 | generate_throw_exception("NullPointerException at call throw_exception", |
duke@435 | 3097 | CAST_FROM_FN_PTR(address, |
duke@435 | 3098 | SharedRuntime:: |
never@3136 | 3099 | throw_NullPointerException_at_call)); |
duke@435 | 3100 | |
duke@435 | 3101 | StubRoutines::_throw_StackOverflowError_entry = |
duke@435 | 3102 | generate_throw_exception("StackOverflowError throw_exception", |
duke@435 | 3103 | CAST_FROM_FN_PTR(address, |
duke@435 | 3104 | SharedRuntime:: |
never@3136 | 3105 | throw_StackOverflowError)); |
duke@435 | 3106 | |
duke@435 | 3107 | // entry points that are platform specific |
never@739 | 3108 | StubRoutines::x86::_f2i_fixup = generate_f2i_fixup(); |
never@739 | 3109 | StubRoutines::x86::_f2l_fixup = generate_f2l_fixup(); |
never@739 | 3110 | StubRoutines::x86::_d2i_fixup = generate_d2i_fixup(); |
never@739 | 3111 | StubRoutines::x86::_d2l_fixup = generate_d2l_fixup(); |
never@739 | 3112 | |
never@739 | 3113 | StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF); |
never@739 | 3114 | StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000); |
never@739 | 3115 | StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF); |
never@739 | 3116 | StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000); |
duke@435 | 3117 | |
duke@435 | 3118 | // support for verify_oop (must happen after universe_init) |
duke@435 | 3119 | StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop(); |
duke@435 | 3120 | |
duke@435 | 3121 | // arraycopy stubs used by compilers |
duke@435 | 3122 | generate_arraycopy_stubs(); |
twisti@1543 | 3123 | |
never@1609 | 3124 | generate_math_stubs(); |
duke@435 | 3125 | } |
duke@435 | 3126 | |
duke@435 | 3127 | public: |
duke@435 | 3128 | StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { |
duke@435 | 3129 | if (all) { |
duke@435 | 3130 | generate_all(); |
duke@435 | 3131 | } else { |
duke@435 | 3132 | generate_initial(); |
duke@435 | 3133 | } |
duke@435 | 3134 | } |
duke@435 | 3135 | }; // end class declaration |
duke@435 | 3136 | |
duke@435 | 3137 | void StubGenerator_generate(CodeBuffer* code, bool all) { |
duke@435 | 3138 | StubGenerator g(code, all); |
duke@435 | 3139 | } |