Wed, 21 Jan 2015 12:38:11 +0100
8068013: [TESTBUG] Aix support in hotspot jtreg tests
Reviewed-by: ctornqvi, fzhinkin, farvidsson
duke@435 | 1 | /* |
drchase@5353 | 2 | * Copyright (c) 2003, 2013, 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" |
twisti@4318 | 26 | #include "asm/macroAssembler.hpp" |
twisti@4318 | 27 | #include "asm/macroAssembler.inline.hpp" |
stefank@2314 | 28 | #include "interpreter/interpreter.hpp" |
stefank@2314 | 29 | #include "nativeInst_x86.hpp" |
stefank@2314 | 30 | #include "oops/instanceOop.hpp" |
coleenp@4037 | 31 | #include "oops/method.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@4299 | 40 | #include "runtime/thread.inline.hpp" |
stefank@2314 | 41 | #include "utilities/top.hpp" |
stefank@2314 | 42 | #ifdef COMPILER2 |
stefank@2314 | 43 | #include "opto/runtime.hpp" |
stefank@2314 | 44 | #endif |
duke@435 | 45 | |
duke@435 | 46 | // Declaration and definition of StubGenerator (no .hpp file). |
duke@435 | 47 | // For a more detailed description of the stub routine structure |
duke@435 | 48 | // see the comment in stubRoutines.hpp |
duke@435 | 49 | |
duke@435 | 50 | #define __ _masm-> |
coleenp@548 | 51 | #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8) |
never@739 | 52 | #define a__ ((Assembler*)_masm)-> |
duke@435 | 53 | |
duke@435 | 54 | #ifdef PRODUCT |
duke@435 | 55 | #define BLOCK_COMMENT(str) /* nothing */ |
duke@435 | 56 | #else |
duke@435 | 57 | #define BLOCK_COMMENT(str) __ block_comment(str) |
duke@435 | 58 | #endif |
duke@435 | 59 | |
duke@435 | 60 | #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") |
duke@435 | 61 | const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions |
duke@435 | 62 | |
duke@435 | 63 | // Stub Code definitions |
duke@435 | 64 | |
duke@435 | 65 | static address handle_unsafe_access() { |
duke@435 | 66 | JavaThread* thread = JavaThread::current(); |
duke@435 | 67 | address pc = thread->saved_exception_pc(); |
duke@435 | 68 | // pc is the instruction which we must emulate |
duke@435 | 69 | // doing a no-op is fine: return garbage from the load |
duke@435 | 70 | // therefore, compute npc |
duke@435 | 71 | address npc = Assembler::locate_next_instruction(pc); |
duke@435 | 72 | |
duke@435 | 73 | // request an async exception |
duke@435 | 74 | thread->set_pending_unsafe_access_error(); |
duke@435 | 75 | |
duke@435 | 76 | // return address of next instruction to execute |
duke@435 | 77 | return npc; |
duke@435 | 78 | } |
duke@435 | 79 | |
duke@435 | 80 | class StubGenerator: public StubCodeGenerator { |
duke@435 | 81 | private: |
duke@435 | 82 | |
duke@435 | 83 | #ifdef PRODUCT |
ccheung@5259 | 84 | #define inc_counter_np(counter) ((void)0) |
duke@435 | 85 | #else |
duke@435 | 86 | void inc_counter_np_(int& counter) { |
never@3314 | 87 | // This can destroy rscratch1 if counter is far from the code cache |
duke@435 | 88 | __ incrementl(ExternalAddress((address)&counter)); |
duke@435 | 89 | } |
duke@435 | 90 | #define inc_counter_np(counter) \ |
duke@435 | 91 | BLOCK_COMMENT("inc_counter " #counter); \ |
duke@435 | 92 | inc_counter_np_(counter); |
duke@435 | 93 | #endif |
duke@435 | 94 | |
duke@435 | 95 | // Call stubs are used to call Java from C |
duke@435 | 96 | // |
duke@435 | 97 | // Linux Arguments: |
duke@435 | 98 | // c_rarg0: call wrapper address address |
duke@435 | 99 | // c_rarg1: result address |
duke@435 | 100 | // c_rarg2: result type BasicType |
coleenp@4037 | 101 | // c_rarg3: method Method* |
duke@435 | 102 | // c_rarg4: (interpreter) entry point address |
duke@435 | 103 | // c_rarg5: parameters intptr_t* |
duke@435 | 104 | // 16(rbp): parameter size (in words) int |
duke@435 | 105 | // 24(rbp): thread Thread* |
duke@435 | 106 | // |
duke@435 | 107 | // [ return_from_Java ] <--- rsp |
duke@435 | 108 | // [ argument word n ] |
duke@435 | 109 | // ... |
duke@435 | 110 | // -12 [ argument word 1 ] |
duke@435 | 111 | // -11 [ saved r15 ] <--- rsp_after_call |
duke@435 | 112 | // -10 [ saved r14 ] |
duke@435 | 113 | // -9 [ saved r13 ] |
duke@435 | 114 | // -8 [ saved r12 ] |
duke@435 | 115 | // -7 [ saved rbx ] |
duke@435 | 116 | // -6 [ call wrapper ] |
duke@435 | 117 | // -5 [ result ] |
duke@435 | 118 | // -4 [ result type ] |
duke@435 | 119 | // -3 [ method ] |
duke@435 | 120 | // -2 [ entry point ] |
duke@435 | 121 | // -1 [ parameters ] |
duke@435 | 122 | // 0 [ saved rbp ] <--- rbp |
duke@435 | 123 | // 1 [ return address ] |
duke@435 | 124 | // 2 [ parameter size ] |
duke@435 | 125 | // 3 [ thread ] |
duke@435 | 126 | // |
duke@435 | 127 | // Windows Arguments: |
duke@435 | 128 | // c_rarg0: call wrapper address address |
duke@435 | 129 | // c_rarg1: result address |
duke@435 | 130 | // c_rarg2: result type BasicType |
coleenp@4037 | 131 | // c_rarg3: method Method* |
duke@435 | 132 | // 48(rbp): (interpreter) entry point address |
duke@435 | 133 | // 56(rbp): parameters intptr_t* |
duke@435 | 134 | // 64(rbp): parameter size (in words) int |
duke@435 | 135 | // 72(rbp): thread Thread* |
duke@435 | 136 | // |
duke@435 | 137 | // [ return_from_Java ] <--- rsp |
duke@435 | 138 | // [ argument word n ] |
duke@435 | 139 | // ... |
iveresov@2689 | 140 | // -28 [ argument word 1 ] |
iveresov@2689 | 141 | // -27 [ saved xmm15 ] <--- rsp_after_call |
iveresov@2689 | 142 | // [ saved xmm7-xmm14 ] |
iveresov@2689 | 143 | // -9 [ saved xmm6 ] (each xmm register takes 2 slots) |
iveresov@2689 | 144 | // -7 [ saved r15 ] |
duke@435 | 145 | // -6 [ saved r14 ] |
duke@435 | 146 | // -5 [ saved r13 ] |
duke@435 | 147 | // -4 [ saved r12 ] |
duke@435 | 148 | // -3 [ saved rdi ] |
duke@435 | 149 | // -2 [ saved rsi ] |
duke@435 | 150 | // -1 [ saved rbx ] |
duke@435 | 151 | // 0 [ saved rbp ] <--- rbp |
duke@435 | 152 | // 1 [ return address ] |
duke@435 | 153 | // 2 [ call wrapper ] |
duke@435 | 154 | // 3 [ result ] |
duke@435 | 155 | // 4 [ result type ] |
duke@435 | 156 | // 5 [ method ] |
duke@435 | 157 | // 6 [ entry point ] |
duke@435 | 158 | // 7 [ parameters ] |
duke@435 | 159 | // 8 [ parameter size ] |
duke@435 | 160 | // 9 [ thread ] |
duke@435 | 161 | // |
duke@435 | 162 | // Windows reserves the callers stack space for arguments 1-4. |
duke@435 | 163 | // We spill c_rarg0-c_rarg3 to this space. |
duke@435 | 164 | |
duke@435 | 165 | // Call stub stack layout word offsets from rbp |
duke@435 | 166 | enum call_stub_layout { |
duke@435 | 167 | #ifdef _WIN64 |
iveresov@2689 | 168 | xmm_save_first = 6, // save from xmm6 |
iveresov@2689 | 169 | xmm_save_last = 15, // to xmm15 |
iveresov@2689 | 170 | xmm_save_base = -9, |
iveresov@2689 | 171 | rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27 |
iveresov@2689 | 172 | r15_off = -7, |
duke@435 | 173 | r14_off = -6, |
duke@435 | 174 | r13_off = -5, |
duke@435 | 175 | r12_off = -4, |
duke@435 | 176 | rdi_off = -3, |
duke@435 | 177 | rsi_off = -2, |
duke@435 | 178 | rbx_off = -1, |
duke@435 | 179 | rbp_off = 0, |
duke@435 | 180 | retaddr_off = 1, |
duke@435 | 181 | call_wrapper_off = 2, |
duke@435 | 182 | result_off = 3, |
duke@435 | 183 | result_type_off = 4, |
duke@435 | 184 | method_off = 5, |
duke@435 | 185 | entry_point_off = 6, |
duke@435 | 186 | parameters_off = 7, |
duke@435 | 187 | parameter_size_off = 8, |
duke@435 | 188 | thread_off = 9 |
duke@435 | 189 | #else |
duke@435 | 190 | rsp_after_call_off = -12, |
duke@435 | 191 | mxcsr_off = rsp_after_call_off, |
duke@435 | 192 | r15_off = -11, |
duke@435 | 193 | r14_off = -10, |
duke@435 | 194 | r13_off = -9, |
duke@435 | 195 | r12_off = -8, |
duke@435 | 196 | rbx_off = -7, |
duke@435 | 197 | call_wrapper_off = -6, |
duke@435 | 198 | result_off = -5, |
duke@435 | 199 | result_type_off = -4, |
duke@435 | 200 | method_off = -3, |
duke@435 | 201 | entry_point_off = -2, |
duke@435 | 202 | parameters_off = -1, |
duke@435 | 203 | rbp_off = 0, |
duke@435 | 204 | retaddr_off = 1, |
duke@435 | 205 | parameter_size_off = 2, |
duke@435 | 206 | thread_off = 3 |
duke@435 | 207 | #endif |
duke@435 | 208 | }; |
duke@435 | 209 | |
iveresov@2689 | 210 | #ifdef _WIN64 |
iveresov@2689 | 211 | Address xmm_save(int reg) { |
iveresov@2689 | 212 | assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range"); |
iveresov@2689 | 213 | return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize); |
iveresov@2689 | 214 | } |
iveresov@2689 | 215 | #endif |
iveresov@2689 | 216 | |
duke@435 | 217 | address generate_call_stub(address& return_address) { |
duke@435 | 218 | assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 && |
duke@435 | 219 | (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off, |
duke@435 | 220 | "adjust this code"); |
duke@435 | 221 | StubCodeMark mark(this, "StubRoutines", "call_stub"); |
duke@435 | 222 | address start = __ pc(); |
duke@435 | 223 | |
duke@435 | 224 | // same as in generate_catch_exception()! |
duke@435 | 225 | const Address rsp_after_call(rbp, rsp_after_call_off * wordSize); |
duke@435 | 226 | |
duke@435 | 227 | const Address call_wrapper (rbp, call_wrapper_off * wordSize); |
duke@435 | 228 | const Address result (rbp, result_off * wordSize); |
duke@435 | 229 | const Address result_type (rbp, result_type_off * wordSize); |
duke@435 | 230 | const Address method (rbp, method_off * wordSize); |
duke@435 | 231 | const Address entry_point (rbp, entry_point_off * wordSize); |
duke@435 | 232 | const Address parameters (rbp, parameters_off * wordSize); |
duke@435 | 233 | const Address parameter_size(rbp, parameter_size_off * wordSize); |
duke@435 | 234 | |
duke@435 | 235 | // same as in generate_catch_exception()! |
duke@435 | 236 | const Address thread (rbp, thread_off * wordSize); |
duke@435 | 237 | |
duke@435 | 238 | const Address r15_save(rbp, r15_off * wordSize); |
duke@435 | 239 | const Address r14_save(rbp, r14_off * wordSize); |
duke@435 | 240 | const Address r13_save(rbp, r13_off * wordSize); |
duke@435 | 241 | const Address r12_save(rbp, r12_off * wordSize); |
duke@435 | 242 | const Address rbx_save(rbp, rbx_off * wordSize); |
duke@435 | 243 | |
duke@435 | 244 | // stub code |
duke@435 | 245 | __ enter(); |
never@739 | 246 | __ subptr(rsp, -rsp_after_call_off * wordSize); |
duke@435 | 247 | |
duke@435 | 248 | // save register parameters |
duke@435 | 249 | #ifndef _WIN64 |
never@739 | 250 | __ movptr(parameters, c_rarg5); // parameters |
never@739 | 251 | __ movptr(entry_point, c_rarg4); // entry_point |
duke@435 | 252 | #endif |
duke@435 | 253 | |
never@739 | 254 | __ movptr(method, c_rarg3); // method |
never@739 | 255 | __ movl(result_type, c_rarg2); // result type |
never@739 | 256 | __ movptr(result, c_rarg1); // result |
never@739 | 257 | __ movptr(call_wrapper, c_rarg0); // call wrapper |
duke@435 | 258 | |
duke@435 | 259 | // save regs belonging to calling function |
never@739 | 260 | __ movptr(rbx_save, rbx); |
never@739 | 261 | __ movptr(r12_save, r12); |
never@739 | 262 | __ movptr(r13_save, r13); |
never@739 | 263 | __ movptr(r14_save, r14); |
never@739 | 264 | __ movptr(r15_save, r15); |
duke@435 | 265 | #ifdef _WIN64 |
iveresov@2689 | 266 | for (int i = 6; i <= 15; i++) { |
iveresov@2689 | 267 | __ movdqu(xmm_save(i), as_XMMRegister(i)); |
iveresov@2689 | 268 | } |
iveresov@2689 | 269 | |
duke@435 | 270 | const Address rdi_save(rbp, rdi_off * wordSize); |
duke@435 | 271 | const Address rsi_save(rbp, rsi_off * wordSize); |
duke@435 | 272 | |
never@739 | 273 | __ movptr(rsi_save, rsi); |
never@739 | 274 | __ movptr(rdi_save, rdi); |
duke@435 | 275 | #else |
duke@435 | 276 | const Address mxcsr_save(rbp, mxcsr_off * wordSize); |
duke@435 | 277 | { |
duke@435 | 278 | Label skip_ldmx; |
duke@435 | 279 | __ stmxcsr(mxcsr_save); |
duke@435 | 280 | __ movl(rax, mxcsr_save); |
duke@435 | 281 | __ andl(rax, MXCSR_MASK); // Only check control and mask bits |
kvn@5439 | 282 | ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); |
duke@435 | 283 | __ cmp32(rax, mxcsr_std); |
duke@435 | 284 | __ jcc(Assembler::equal, skip_ldmx); |
duke@435 | 285 | __ ldmxcsr(mxcsr_std); |
duke@435 | 286 | __ bind(skip_ldmx); |
duke@435 | 287 | } |
duke@435 | 288 | #endif |
duke@435 | 289 | |
duke@435 | 290 | // Load up thread register |
never@739 | 291 | __ movptr(r15_thread, thread); |
coleenp@548 | 292 | __ reinit_heapbase(); |
duke@435 | 293 | |
duke@435 | 294 | #ifdef ASSERT |
duke@435 | 295 | // make sure we have no pending exceptions |
duke@435 | 296 | { |
duke@435 | 297 | Label L; |
never@739 | 298 | __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
duke@435 | 299 | __ jcc(Assembler::equal, L); |
duke@435 | 300 | __ stop("StubRoutines::call_stub: entered with pending exception"); |
duke@435 | 301 | __ bind(L); |
duke@435 | 302 | } |
duke@435 | 303 | #endif |
duke@435 | 304 | |
duke@435 | 305 | // pass parameters if any |
duke@435 | 306 | BLOCK_COMMENT("pass parameters if any"); |
duke@435 | 307 | Label parameters_done; |
duke@435 | 308 | __ movl(c_rarg3, parameter_size); |
duke@435 | 309 | __ testl(c_rarg3, c_rarg3); |
duke@435 | 310 | __ jcc(Assembler::zero, parameters_done); |
duke@435 | 311 | |
duke@435 | 312 | Label loop; |
never@739 | 313 | __ movptr(c_rarg2, parameters); // parameter pointer |
never@739 | 314 | __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1 |
duke@435 | 315 | __ BIND(loop); |
never@739 | 316 | __ movptr(rax, Address(c_rarg2, 0));// get parameter |
never@739 | 317 | __ addptr(c_rarg2, wordSize); // advance to next parameter |
never@739 | 318 | __ decrementl(c_rarg1); // decrement counter |
never@739 | 319 | __ push(rax); // pass parameter |
duke@435 | 320 | __ jcc(Assembler::notZero, loop); |
duke@435 | 321 | |
duke@435 | 322 | // call Java function |
duke@435 | 323 | __ BIND(parameters_done); |
coleenp@4037 | 324 | __ movptr(rbx, method); // get Method* |
never@739 | 325 | __ movptr(c_rarg1, entry_point); // get entry_point |
never@739 | 326 | __ mov(r13, rsp); // set sender sp |
duke@435 | 327 | BLOCK_COMMENT("call Java function"); |
duke@435 | 328 | __ call(c_rarg1); |
duke@435 | 329 | |
duke@435 | 330 | BLOCK_COMMENT("call_stub_return_address:"); |
duke@435 | 331 | return_address = __ pc(); |
duke@435 | 332 | |
duke@435 | 333 | // store result depending on type (everything that is not |
duke@435 | 334 | // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT) |
never@739 | 335 | __ movptr(c_rarg0, result); |
duke@435 | 336 | Label is_long, is_float, is_double, exit; |
duke@435 | 337 | __ movl(c_rarg1, result_type); |
duke@435 | 338 | __ cmpl(c_rarg1, T_OBJECT); |
duke@435 | 339 | __ jcc(Assembler::equal, is_long); |
duke@435 | 340 | __ cmpl(c_rarg1, T_LONG); |
duke@435 | 341 | __ jcc(Assembler::equal, is_long); |
duke@435 | 342 | __ cmpl(c_rarg1, T_FLOAT); |
duke@435 | 343 | __ jcc(Assembler::equal, is_float); |
duke@435 | 344 | __ cmpl(c_rarg1, T_DOUBLE); |
duke@435 | 345 | __ jcc(Assembler::equal, is_double); |
duke@435 | 346 | |
duke@435 | 347 | // handle T_INT case |
duke@435 | 348 | __ movl(Address(c_rarg0, 0), rax); |
duke@435 | 349 | |
duke@435 | 350 | __ BIND(exit); |
duke@435 | 351 | |
duke@435 | 352 | // pop parameters |
never@739 | 353 | __ lea(rsp, rsp_after_call); |
duke@435 | 354 | |
duke@435 | 355 | #ifdef ASSERT |
duke@435 | 356 | // verify that threads correspond |
duke@435 | 357 | { |
duke@435 | 358 | Label L, S; |
never@739 | 359 | __ cmpptr(r15_thread, thread); |
duke@435 | 360 | __ jcc(Assembler::notEqual, S); |
duke@435 | 361 | __ get_thread(rbx); |
never@739 | 362 | __ cmpptr(r15_thread, rbx); |
duke@435 | 363 | __ jcc(Assembler::equal, L); |
duke@435 | 364 | __ bind(S); |
duke@435 | 365 | __ jcc(Assembler::equal, L); |
duke@435 | 366 | __ stop("StubRoutines::call_stub: threads must correspond"); |
duke@435 | 367 | __ bind(L); |
duke@435 | 368 | } |
duke@435 | 369 | #endif |
duke@435 | 370 | |
duke@435 | 371 | // restore regs belonging to calling function |
iveresov@2689 | 372 | #ifdef _WIN64 |
iveresov@2689 | 373 | for (int i = 15; i >= 6; i--) { |
iveresov@2689 | 374 | __ movdqu(as_XMMRegister(i), xmm_save(i)); |
iveresov@2689 | 375 | } |
iveresov@2689 | 376 | #endif |
never@739 | 377 | __ movptr(r15, r15_save); |
never@739 | 378 | __ movptr(r14, r14_save); |
never@739 | 379 | __ movptr(r13, r13_save); |
never@739 | 380 | __ movptr(r12, r12_save); |
never@739 | 381 | __ movptr(rbx, rbx_save); |
duke@435 | 382 | |
duke@435 | 383 | #ifdef _WIN64 |
never@739 | 384 | __ movptr(rdi, rdi_save); |
never@739 | 385 | __ movptr(rsi, rsi_save); |
duke@435 | 386 | #else |
duke@435 | 387 | __ ldmxcsr(mxcsr_save); |
duke@435 | 388 | #endif |
duke@435 | 389 | |
duke@435 | 390 | // restore rsp |
never@739 | 391 | __ addptr(rsp, -rsp_after_call_off * wordSize); |
duke@435 | 392 | |
duke@435 | 393 | // return |
never@739 | 394 | __ pop(rbp); |
duke@435 | 395 | __ ret(0); |
duke@435 | 396 | |
duke@435 | 397 | // handle return types different from T_INT |
duke@435 | 398 | __ BIND(is_long); |
duke@435 | 399 | __ movq(Address(c_rarg0, 0), rax); |
duke@435 | 400 | __ jmp(exit); |
duke@435 | 401 | |
duke@435 | 402 | __ BIND(is_float); |
duke@435 | 403 | __ movflt(Address(c_rarg0, 0), xmm0); |
duke@435 | 404 | __ jmp(exit); |
duke@435 | 405 | |
duke@435 | 406 | __ BIND(is_double); |
duke@435 | 407 | __ movdbl(Address(c_rarg0, 0), xmm0); |
duke@435 | 408 | __ jmp(exit); |
duke@435 | 409 | |
duke@435 | 410 | return start; |
duke@435 | 411 | } |
duke@435 | 412 | |
duke@435 | 413 | // Return point for a Java call if there's an exception thrown in |
duke@435 | 414 | // Java code. The exception is caught and transformed into a |
duke@435 | 415 | // pending exception stored in JavaThread that can be tested from |
duke@435 | 416 | // within the VM. |
duke@435 | 417 | // |
duke@435 | 418 | // Note: Usually the parameters are removed by the callee. In case |
duke@435 | 419 | // of an exception crossing an activation frame boundary, that is |
duke@435 | 420 | // not the case if the callee is compiled code => need to setup the |
duke@435 | 421 | // rsp. |
duke@435 | 422 | // |
duke@435 | 423 | // rax: exception oop |
duke@435 | 424 | |
duke@435 | 425 | address generate_catch_exception() { |
duke@435 | 426 | StubCodeMark mark(this, "StubRoutines", "catch_exception"); |
duke@435 | 427 | address start = __ pc(); |
duke@435 | 428 | |
duke@435 | 429 | // same as in generate_call_stub(): |
duke@435 | 430 | const Address rsp_after_call(rbp, rsp_after_call_off * wordSize); |
duke@435 | 431 | const Address thread (rbp, thread_off * wordSize); |
duke@435 | 432 | |
duke@435 | 433 | #ifdef ASSERT |
duke@435 | 434 | // verify that threads correspond |
duke@435 | 435 | { |
duke@435 | 436 | Label L, S; |
never@739 | 437 | __ cmpptr(r15_thread, thread); |
duke@435 | 438 | __ jcc(Assembler::notEqual, S); |
duke@435 | 439 | __ get_thread(rbx); |
never@739 | 440 | __ cmpptr(r15_thread, rbx); |
duke@435 | 441 | __ jcc(Assembler::equal, L); |
duke@435 | 442 | __ bind(S); |
duke@435 | 443 | __ stop("StubRoutines::catch_exception: threads must correspond"); |
duke@435 | 444 | __ bind(L); |
duke@435 | 445 | } |
duke@435 | 446 | #endif |
duke@435 | 447 | |
duke@435 | 448 | // set pending exception |
duke@435 | 449 | __ verify_oop(rax); |
duke@435 | 450 | |
never@739 | 451 | __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax); |
duke@435 | 452 | __ lea(rscratch1, ExternalAddress((address)__FILE__)); |
never@739 | 453 | __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1); |
duke@435 | 454 | __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__); |
duke@435 | 455 | |
duke@435 | 456 | // complete return to VM |
duke@435 | 457 | assert(StubRoutines::_call_stub_return_address != NULL, |
duke@435 | 458 | "_call_stub_return_address must have been generated before"); |
duke@435 | 459 | __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address)); |
duke@435 | 460 | |
duke@435 | 461 | return start; |
duke@435 | 462 | } |
duke@435 | 463 | |
duke@435 | 464 | // Continuation point for runtime calls returning with a pending |
duke@435 | 465 | // exception. The pending exception check happened in the runtime |
duke@435 | 466 | // or native call stub. The pending exception in Thread is |
duke@435 | 467 | // converted into a Java-level exception. |
duke@435 | 468 | // |
duke@435 | 469 | // Contract with Java-level exception handlers: |
duke@435 | 470 | // rax: exception |
duke@435 | 471 | // rdx: throwing pc |
duke@435 | 472 | // |
duke@435 | 473 | // NOTE: At entry of this stub, exception-pc must be on stack !! |
duke@435 | 474 | |
duke@435 | 475 | address generate_forward_exception() { |
duke@435 | 476 | StubCodeMark mark(this, "StubRoutines", "forward exception"); |
duke@435 | 477 | address start = __ pc(); |
duke@435 | 478 | |
duke@435 | 479 | // Upon entry, the sp points to the return address returning into |
duke@435 | 480 | // Java (interpreted or compiled) code; i.e., the return address |
duke@435 | 481 | // becomes the throwing pc. |
duke@435 | 482 | // |
duke@435 | 483 | // Arguments pushed before the runtime call are still on the stack |
duke@435 | 484 | // but the exception handler will reset the stack pointer -> |
duke@435 | 485 | // ignore them. A potential result in registers can be ignored as |
duke@435 | 486 | // well. |
duke@435 | 487 | |
duke@435 | 488 | #ifdef ASSERT |
duke@435 | 489 | // make sure this code is only executed if there is a pending exception |
duke@435 | 490 | { |
duke@435 | 491 | Label L; |
never@739 | 492 | __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL); |
duke@435 | 493 | __ jcc(Assembler::notEqual, L); |
duke@435 | 494 | __ stop("StubRoutines::forward exception: no pending exception (1)"); |
duke@435 | 495 | __ bind(L); |
duke@435 | 496 | } |
duke@435 | 497 | #endif |
duke@435 | 498 | |
duke@435 | 499 | // compute exception handler into rbx |
never@739 | 500 | __ movptr(c_rarg0, Address(rsp, 0)); |
duke@435 | 501 | BLOCK_COMMENT("call exception_handler_for_return_address"); |
duke@435 | 502 | __ call_VM_leaf(CAST_FROM_FN_PTR(address, |
duke@435 | 503 | SharedRuntime::exception_handler_for_return_address), |
twisti@1730 | 504 | r15_thread, c_rarg0); |
never@739 | 505 | __ mov(rbx, rax); |
duke@435 | 506 | |
duke@435 | 507 | // setup rax & rdx, remove return address & clear pending exception |
never@739 | 508 | __ pop(rdx); |
never@739 | 509 | __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset())); |
xlu@947 | 510 | __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
duke@435 | 511 | |
duke@435 | 512 | #ifdef ASSERT |
duke@435 | 513 | // make sure exception is set |
duke@435 | 514 | { |
duke@435 | 515 | Label L; |
never@739 | 516 | __ testptr(rax, rax); |
duke@435 | 517 | __ jcc(Assembler::notEqual, L); |
duke@435 | 518 | __ stop("StubRoutines::forward exception: no pending exception (2)"); |
duke@435 | 519 | __ bind(L); |
duke@435 | 520 | } |
duke@435 | 521 | #endif |
duke@435 | 522 | |
duke@435 | 523 | // continue at exception handler (return address removed) |
duke@435 | 524 | // rax: exception |
duke@435 | 525 | // rbx: exception handler |
duke@435 | 526 | // rdx: throwing pc |
duke@435 | 527 | __ verify_oop(rax); |
duke@435 | 528 | __ jmp(rbx); |
duke@435 | 529 | |
duke@435 | 530 | return start; |
duke@435 | 531 | } |
duke@435 | 532 | |
duke@435 | 533 | // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest) |
duke@435 | 534 | // |
duke@435 | 535 | // Arguments : |
duke@435 | 536 | // c_rarg0: exchange_value |
duke@435 | 537 | // c_rarg0: dest |
duke@435 | 538 | // |
duke@435 | 539 | // Result: |
duke@435 | 540 | // *dest <- ex, return (orig *dest) |
duke@435 | 541 | address generate_atomic_xchg() { |
duke@435 | 542 | StubCodeMark mark(this, "StubRoutines", "atomic_xchg"); |
duke@435 | 543 | address start = __ pc(); |
duke@435 | 544 | |
duke@435 | 545 | __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow |
duke@435 | 546 | __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK |
duke@435 | 547 | __ ret(0); |
duke@435 | 548 | |
duke@435 | 549 | return start; |
duke@435 | 550 | } |
duke@435 | 551 | |
duke@435 | 552 | // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest) |
duke@435 | 553 | // |
duke@435 | 554 | // Arguments : |
duke@435 | 555 | // c_rarg0: exchange_value |
duke@435 | 556 | // c_rarg1: dest |
duke@435 | 557 | // |
duke@435 | 558 | // Result: |
duke@435 | 559 | // *dest <- ex, return (orig *dest) |
duke@435 | 560 | address generate_atomic_xchg_ptr() { |
duke@435 | 561 | StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr"); |
duke@435 | 562 | address start = __ pc(); |
duke@435 | 563 | |
never@739 | 564 | __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow |
never@739 | 565 | __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK |
duke@435 | 566 | __ ret(0); |
duke@435 | 567 | |
duke@435 | 568 | return start; |
duke@435 | 569 | } |
duke@435 | 570 | |
duke@435 | 571 | // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest, |
duke@435 | 572 | // jint compare_value) |
duke@435 | 573 | // |
duke@435 | 574 | // Arguments : |
duke@435 | 575 | // c_rarg0: exchange_value |
duke@435 | 576 | // c_rarg1: dest |
duke@435 | 577 | // c_rarg2: compare_value |
duke@435 | 578 | // |
duke@435 | 579 | // Result: |
duke@435 | 580 | // if ( compare_value == *dest ) { |
duke@435 | 581 | // *dest = exchange_value |
duke@435 | 582 | // return compare_value; |
duke@435 | 583 | // else |
duke@435 | 584 | // return *dest; |
duke@435 | 585 | address generate_atomic_cmpxchg() { |
duke@435 | 586 | StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg"); |
duke@435 | 587 | address start = __ pc(); |
duke@435 | 588 | |
duke@435 | 589 | __ movl(rax, c_rarg2); |
duke@435 | 590 | if ( os::is_MP() ) __ lock(); |
duke@435 | 591 | __ cmpxchgl(c_rarg0, Address(c_rarg1, 0)); |
duke@435 | 592 | __ ret(0); |
duke@435 | 593 | |
duke@435 | 594 | return start; |
duke@435 | 595 | } |
duke@435 | 596 | |
duke@435 | 597 | // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value, |
duke@435 | 598 | // volatile jlong* dest, |
duke@435 | 599 | // jlong compare_value) |
duke@435 | 600 | // Arguments : |
duke@435 | 601 | // c_rarg0: exchange_value |
duke@435 | 602 | // c_rarg1: dest |
duke@435 | 603 | // c_rarg2: compare_value |
duke@435 | 604 | // |
duke@435 | 605 | // Result: |
duke@435 | 606 | // if ( compare_value == *dest ) { |
duke@435 | 607 | // *dest = exchange_value |
duke@435 | 608 | // return compare_value; |
duke@435 | 609 | // else |
duke@435 | 610 | // return *dest; |
duke@435 | 611 | address generate_atomic_cmpxchg_long() { |
duke@435 | 612 | StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long"); |
duke@435 | 613 | address start = __ pc(); |
duke@435 | 614 | |
duke@435 | 615 | __ movq(rax, c_rarg2); |
duke@435 | 616 | if ( os::is_MP() ) __ lock(); |
duke@435 | 617 | __ cmpxchgq(c_rarg0, Address(c_rarg1, 0)); |
duke@435 | 618 | __ ret(0); |
duke@435 | 619 | |
duke@435 | 620 | return start; |
duke@435 | 621 | } |
duke@435 | 622 | |
duke@435 | 623 | // Support for jint atomic::add(jint add_value, volatile jint* dest) |
duke@435 | 624 | // |
duke@435 | 625 | // Arguments : |
duke@435 | 626 | // c_rarg0: add_value |
duke@435 | 627 | // c_rarg1: dest |
duke@435 | 628 | // |
duke@435 | 629 | // Result: |
duke@435 | 630 | // *dest += add_value |
duke@435 | 631 | // return *dest; |
duke@435 | 632 | address generate_atomic_add() { |
duke@435 | 633 | StubCodeMark mark(this, "StubRoutines", "atomic_add"); |
duke@435 | 634 | address start = __ pc(); |
duke@435 | 635 | |
duke@435 | 636 | __ movl(rax, c_rarg0); |
duke@435 | 637 | if ( os::is_MP() ) __ lock(); |
duke@435 | 638 | __ xaddl(Address(c_rarg1, 0), c_rarg0); |
duke@435 | 639 | __ addl(rax, c_rarg0); |
duke@435 | 640 | __ ret(0); |
duke@435 | 641 | |
duke@435 | 642 | return start; |
duke@435 | 643 | } |
duke@435 | 644 | |
duke@435 | 645 | // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest) |
duke@435 | 646 | // |
duke@435 | 647 | // Arguments : |
duke@435 | 648 | // c_rarg0: add_value |
duke@435 | 649 | // c_rarg1: dest |
duke@435 | 650 | // |
duke@435 | 651 | // Result: |
duke@435 | 652 | // *dest += add_value |
duke@435 | 653 | // return *dest; |
duke@435 | 654 | address generate_atomic_add_ptr() { |
duke@435 | 655 | StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr"); |
duke@435 | 656 | address start = __ pc(); |
duke@435 | 657 | |
never@739 | 658 | __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow |
duke@435 | 659 | if ( os::is_MP() ) __ lock(); |
never@739 | 660 | __ xaddptr(Address(c_rarg1, 0), c_rarg0); |
never@739 | 661 | __ addptr(rax, c_rarg0); |
duke@435 | 662 | __ ret(0); |
duke@435 | 663 | |
duke@435 | 664 | return start; |
duke@435 | 665 | } |
duke@435 | 666 | |
duke@435 | 667 | // Support for intptr_t OrderAccess::fence() |
duke@435 | 668 | // |
duke@435 | 669 | // Arguments : |
duke@435 | 670 | // |
duke@435 | 671 | // Result: |
duke@435 | 672 | address generate_orderaccess_fence() { |
duke@435 | 673 | StubCodeMark mark(this, "StubRoutines", "orderaccess_fence"); |
duke@435 | 674 | address start = __ pc(); |
never@1106 | 675 | __ membar(Assembler::StoreLoad); |
duke@435 | 676 | __ ret(0); |
duke@435 | 677 | |
duke@435 | 678 | return start; |
duke@435 | 679 | } |
duke@435 | 680 | |
duke@435 | 681 | // Support for intptr_t get_previous_fp() |
duke@435 | 682 | // |
duke@435 | 683 | // This routine is used to find the previous frame pointer for the |
duke@435 | 684 | // caller (current_frame_guess). This is used as part of debugging |
duke@435 | 685 | // ps() is seemingly lost trying to find frames. |
duke@435 | 686 | // This code assumes that caller current_frame_guess) has a frame. |
duke@435 | 687 | address generate_get_previous_fp() { |
duke@435 | 688 | StubCodeMark mark(this, "StubRoutines", "get_previous_fp"); |
duke@435 | 689 | const Address old_fp(rbp, 0); |
duke@435 | 690 | const Address older_fp(rax, 0); |
duke@435 | 691 | address start = __ pc(); |
duke@435 | 692 | |
duke@435 | 693 | __ enter(); |
never@739 | 694 | __ movptr(rax, old_fp); // callers fp |
never@739 | 695 | __ movptr(rax, older_fp); // the frame for ps() |
never@739 | 696 | __ pop(rbp); |
duke@435 | 697 | __ ret(0); |
duke@435 | 698 | |
duke@435 | 699 | return start; |
duke@435 | 700 | } |
duke@435 | 701 | |
roland@3606 | 702 | // Support for intptr_t get_previous_sp() |
roland@3606 | 703 | // |
roland@3606 | 704 | // This routine is used to find the previous stack pointer for the |
roland@3606 | 705 | // caller. |
roland@3606 | 706 | address generate_get_previous_sp() { |
roland@3606 | 707 | StubCodeMark mark(this, "StubRoutines", "get_previous_sp"); |
roland@3606 | 708 | address start = __ pc(); |
roland@3606 | 709 | |
roland@3606 | 710 | __ movptr(rax, rsp); |
roland@3606 | 711 | __ addptr(rax, 8); // return address is at the top of the stack. |
roland@3606 | 712 | __ ret(0); |
roland@3606 | 713 | |
roland@3606 | 714 | return start; |
roland@3606 | 715 | } |
roland@3606 | 716 | |
duke@435 | 717 | //---------------------------------------------------------------------------------------------------- |
duke@435 | 718 | // Support for void verify_mxcsr() |
duke@435 | 719 | // |
duke@435 | 720 | // This routine is used with -Xcheck:jni to verify that native |
duke@435 | 721 | // JNI code does not return to Java code without restoring the |
duke@435 | 722 | // MXCSR register to our expected state. |
duke@435 | 723 | |
duke@435 | 724 | address generate_verify_mxcsr() { |
duke@435 | 725 | StubCodeMark mark(this, "StubRoutines", "verify_mxcsr"); |
duke@435 | 726 | address start = __ pc(); |
duke@435 | 727 | |
duke@435 | 728 | const Address mxcsr_save(rsp, 0); |
duke@435 | 729 | |
duke@435 | 730 | if (CheckJNICalls) { |
duke@435 | 731 | Label ok_ret; |
kvn@5439 | 732 | ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); |
never@739 | 733 | __ push(rax); |
never@739 | 734 | __ subptr(rsp, wordSize); // allocate a temp location |
duke@435 | 735 | __ stmxcsr(mxcsr_save); |
duke@435 | 736 | __ movl(rax, mxcsr_save); |
duke@435 | 737 | __ andl(rax, MXCSR_MASK); // Only check control and mask bits |
kvn@5439 | 738 | __ cmp32(rax, mxcsr_std); |
duke@435 | 739 | __ jcc(Assembler::equal, ok_ret); |
duke@435 | 740 | |
duke@435 | 741 | __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall"); |
duke@435 | 742 | |
kvn@5439 | 743 | __ ldmxcsr(mxcsr_std); |
duke@435 | 744 | |
duke@435 | 745 | __ bind(ok_ret); |
never@739 | 746 | __ addptr(rsp, wordSize); |
never@739 | 747 | __ pop(rax); |
duke@435 | 748 | } |
duke@435 | 749 | |
duke@435 | 750 | __ ret(0); |
duke@435 | 751 | |
duke@435 | 752 | return start; |
duke@435 | 753 | } |
duke@435 | 754 | |
duke@435 | 755 | address generate_f2i_fixup() { |
duke@435 | 756 | StubCodeMark mark(this, "StubRoutines", "f2i_fixup"); |
duke@435 | 757 | Address inout(rsp, 5 * wordSize); // return address + 4 saves |
duke@435 | 758 | |
duke@435 | 759 | address start = __ pc(); |
duke@435 | 760 | |
duke@435 | 761 | Label L; |
duke@435 | 762 | |
never@739 | 763 | __ push(rax); |
never@739 | 764 | __ push(c_rarg3); |
never@739 | 765 | __ push(c_rarg2); |
never@739 | 766 | __ push(c_rarg1); |
duke@435 | 767 | |
duke@435 | 768 | __ movl(rax, 0x7f800000); |
duke@435 | 769 | __ xorl(c_rarg3, c_rarg3); |
duke@435 | 770 | __ movl(c_rarg2, inout); |
duke@435 | 771 | __ movl(c_rarg1, c_rarg2); |
duke@435 | 772 | __ andl(c_rarg1, 0x7fffffff); |
duke@435 | 773 | __ cmpl(rax, c_rarg1); // NaN? -> 0 |
duke@435 | 774 | __ jcc(Assembler::negative, L); |
duke@435 | 775 | __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint |
duke@435 | 776 | __ movl(c_rarg3, 0x80000000); |
duke@435 | 777 | __ movl(rax, 0x7fffffff); |
duke@435 | 778 | __ cmovl(Assembler::positive, c_rarg3, rax); |
duke@435 | 779 | |
duke@435 | 780 | __ bind(L); |
never@739 | 781 | __ movptr(inout, c_rarg3); |
never@739 | 782 | |
never@739 | 783 | __ pop(c_rarg1); |
never@739 | 784 | __ pop(c_rarg2); |
never@739 | 785 | __ pop(c_rarg3); |
never@739 | 786 | __ pop(rax); |
duke@435 | 787 | |
duke@435 | 788 | __ ret(0); |
duke@435 | 789 | |
duke@435 | 790 | return start; |
duke@435 | 791 | } |
duke@435 | 792 | |
duke@435 | 793 | address generate_f2l_fixup() { |
duke@435 | 794 | StubCodeMark mark(this, "StubRoutines", "f2l_fixup"); |
duke@435 | 795 | Address inout(rsp, 5 * wordSize); // return address + 4 saves |
duke@435 | 796 | address start = __ pc(); |
duke@435 | 797 | |
duke@435 | 798 | Label L; |
duke@435 | 799 | |
never@739 | 800 | __ push(rax); |
never@739 | 801 | __ push(c_rarg3); |
never@739 | 802 | __ push(c_rarg2); |
never@739 | 803 | __ push(c_rarg1); |
duke@435 | 804 | |
duke@435 | 805 | __ movl(rax, 0x7f800000); |
duke@435 | 806 | __ xorl(c_rarg3, c_rarg3); |
duke@435 | 807 | __ movl(c_rarg2, inout); |
duke@435 | 808 | __ movl(c_rarg1, c_rarg2); |
duke@435 | 809 | __ andl(c_rarg1, 0x7fffffff); |
duke@435 | 810 | __ cmpl(rax, c_rarg1); // NaN? -> 0 |
duke@435 | 811 | __ jcc(Assembler::negative, L); |
duke@435 | 812 | __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong |
duke@435 | 813 | __ mov64(c_rarg3, 0x8000000000000000); |
duke@435 | 814 | __ mov64(rax, 0x7fffffffffffffff); |
never@739 | 815 | __ cmov(Assembler::positive, c_rarg3, rax); |
duke@435 | 816 | |
duke@435 | 817 | __ bind(L); |
never@739 | 818 | __ movptr(inout, c_rarg3); |
never@739 | 819 | |
never@739 | 820 | __ pop(c_rarg1); |
never@739 | 821 | __ pop(c_rarg2); |
never@739 | 822 | __ pop(c_rarg3); |
never@739 | 823 | __ pop(rax); |
duke@435 | 824 | |
duke@435 | 825 | __ ret(0); |
duke@435 | 826 | |
duke@435 | 827 | return start; |
duke@435 | 828 | } |
duke@435 | 829 | |
duke@435 | 830 | address generate_d2i_fixup() { |
duke@435 | 831 | StubCodeMark mark(this, "StubRoutines", "d2i_fixup"); |
duke@435 | 832 | Address inout(rsp, 6 * wordSize); // return address + 5 saves |
duke@435 | 833 | |
duke@435 | 834 | address start = __ pc(); |
duke@435 | 835 | |
duke@435 | 836 | Label L; |
duke@435 | 837 | |
never@739 | 838 | __ push(rax); |
never@739 | 839 | __ push(c_rarg3); |
never@739 | 840 | __ push(c_rarg2); |
never@739 | 841 | __ push(c_rarg1); |
never@739 | 842 | __ push(c_rarg0); |
duke@435 | 843 | |
duke@435 | 844 | __ movl(rax, 0x7ff00000); |
duke@435 | 845 | __ movq(c_rarg2, inout); |
duke@435 | 846 | __ movl(c_rarg3, c_rarg2); |
never@739 | 847 | __ mov(c_rarg1, c_rarg2); |
never@739 | 848 | __ mov(c_rarg0, c_rarg2); |
duke@435 | 849 | __ negl(c_rarg3); |
never@739 | 850 | __ shrptr(c_rarg1, 0x20); |
duke@435 | 851 | __ orl(c_rarg3, c_rarg2); |
duke@435 | 852 | __ andl(c_rarg1, 0x7fffffff); |
duke@435 | 853 | __ xorl(c_rarg2, c_rarg2); |
duke@435 | 854 | __ shrl(c_rarg3, 0x1f); |
duke@435 | 855 | __ orl(c_rarg1, c_rarg3); |
duke@435 | 856 | __ cmpl(rax, c_rarg1); |
duke@435 | 857 | __ jcc(Assembler::negative, L); // NaN -> 0 |
never@739 | 858 | __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint |
duke@435 | 859 | __ movl(c_rarg2, 0x80000000); |
duke@435 | 860 | __ movl(rax, 0x7fffffff); |
never@739 | 861 | __ cmov(Assembler::positive, c_rarg2, rax); |
duke@435 | 862 | |
duke@435 | 863 | __ bind(L); |
never@739 | 864 | __ movptr(inout, c_rarg2); |
never@739 | 865 | |
never@739 | 866 | __ pop(c_rarg0); |
never@739 | 867 | __ pop(c_rarg1); |
never@739 | 868 | __ pop(c_rarg2); |
never@739 | 869 | __ pop(c_rarg3); |
never@739 | 870 | __ pop(rax); |
duke@435 | 871 | |
duke@435 | 872 | __ ret(0); |
duke@435 | 873 | |
duke@435 | 874 | return start; |
duke@435 | 875 | } |
duke@435 | 876 | |
duke@435 | 877 | address generate_d2l_fixup() { |
duke@435 | 878 | StubCodeMark mark(this, "StubRoutines", "d2l_fixup"); |
duke@435 | 879 | Address inout(rsp, 6 * wordSize); // return address + 5 saves |
duke@435 | 880 | |
duke@435 | 881 | address start = __ pc(); |
duke@435 | 882 | |
duke@435 | 883 | Label L; |
duke@435 | 884 | |
never@739 | 885 | __ push(rax); |
never@739 | 886 | __ push(c_rarg3); |
never@739 | 887 | __ push(c_rarg2); |
never@739 | 888 | __ push(c_rarg1); |
never@739 | 889 | __ push(c_rarg0); |
duke@435 | 890 | |
duke@435 | 891 | __ movl(rax, 0x7ff00000); |
duke@435 | 892 | __ movq(c_rarg2, inout); |
duke@435 | 893 | __ movl(c_rarg3, c_rarg2); |
never@739 | 894 | __ mov(c_rarg1, c_rarg2); |
never@739 | 895 | __ mov(c_rarg0, c_rarg2); |
duke@435 | 896 | __ negl(c_rarg3); |
never@739 | 897 | __ shrptr(c_rarg1, 0x20); |
duke@435 | 898 | __ orl(c_rarg3, c_rarg2); |
duke@435 | 899 | __ andl(c_rarg1, 0x7fffffff); |
duke@435 | 900 | __ xorl(c_rarg2, c_rarg2); |
duke@435 | 901 | __ shrl(c_rarg3, 0x1f); |
duke@435 | 902 | __ orl(c_rarg1, c_rarg3); |
duke@435 | 903 | __ cmpl(rax, c_rarg1); |
duke@435 | 904 | __ jcc(Assembler::negative, L); // NaN -> 0 |
duke@435 | 905 | __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong |
duke@435 | 906 | __ mov64(c_rarg2, 0x8000000000000000); |
duke@435 | 907 | __ mov64(rax, 0x7fffffffffffffff); |
duke@435 | 908 | __ cmovq(Assembler::positive, c_rarg2, rax); |
duke@435 | 909 | |
duke@435 | 910 | __ bind(L); |
duke@435 | 911 | __ movq(inout, c_rarg2); |
duke@435 | 912 | |
never@739 | 913 | __ pop(c_rarg0); |
never@739 | 914 | __ pop(c_rarg1); |
never@739 | 915 | __ pop(c_rarg2); |
never@739 | 916 | __ pop(c_rarg3); |
never@739 | 917 | __ pop(rax); |
duke@435 | 918 | |
duke@435 | 919 | __ ret(0); |
duke@435 | 920 | |
duke@435 | 921 | return start; |
duke@435 | 922 | } |
duke@435 | 923 | |
duke@435 | 924 | address generate_fp_mask(const char *stub_name, int64_t mask) { |
kvn@1800 | 925 | __ align(CodeEntryAlignment); |
duke@435 | 926 | StubCodeMark mark(this, "StubRoutines", stub_name); |
duke@435 | 927 | address start = __ pc(); |
duke@435 | 928 | |
duke@435 | 929 | __ emit_data64( mask, relocInfo::none ); |
duke@435 | 930 | __ emit_data64( mask, relocInfo::none ); |
duke@435 | 931 | |
duke@435 | 932 | return start; |
duke@435 | 933 | } |
duke@435 | 934 | |
duke@435 | 935 | // The following routine generates a subroutine to throw an |
duke@435 | 936 | // asynchronous UnknownError when an unsafe access gets a fault that |
duke@435 | 937 | // could not be reasonably prevented by the programmer. (Example: |
duke@435 | 938 | // SIGBUS/OBJERR.) |
duke@435 | 939 | address generate_handler_for_unsafe_access() { |
duke@435 | 940 | StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access"); |
duke@435 | 941 | address start = __ pc(); |
duke@435 | 942 | |
never@739 | 943 | __ push(0); // hole for return address-to-be |
never@739 | 944 | __ pusha(); // push registers |
duke@435 | 945 | Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord); |
duke@435 | 946 | |
never@3136 | 947 | // FIXME: this probably needs alignment logic |
never@3136 | 948 | |
never@739 | 949 | __ subptr(rsp, frame::arg_reg_save_area_bytes); |
duke@435 | 950 | BLOCK_COMMENT("call handle_unsafe_access"); |
duke@435 | 951 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access))); |
never@739 | 952 | __ addptr(rsp, frame::arg_reg_save_area_bytes); |
never@739 | 953 | |
never@739 | 954 | __ movptr(next_pc, rax); // stuff next address |
never@739 | 955 | __ popa(); |
duke@435 | 956 | __ ret(0); // jump to next address |
duke@435 | 957 | |
duke@435 | 958 | return start; |
duke@435 | 959 | } |
duke@435 | 960 | |
duke@435 | 961 | // Non-destructive plausibility checks for oops |
duke@435 | 962 | // |
duke@435 | 963 | // Arguments: |
duke@435 | 964 | // all args on stack! |
duke@435 | 965 | // |
duke@435 | 966 | // Stack after saving c_rarg3: |
duke@435 | 967 | // [tos + 0]: saved c_rarg3 |
duke@435 | 968 | // [tos + 1]: saved c_rarg2 |
kvn@559 | 969 | // [tos + 2]: saved r12 (several TemplateTable methods use it) |
kvn@559 | 970 | // [tos + 3]: saved flags |
kvn@559 | 971 | // [tos + 4]: return address |
kvn@559 | 972 | // * [tos + 5]: error message (char*) |
kvn@559 | 973 | // * [tos + 6]: object to verify (oop) |
kvn@559 | 974 | // * [tos + 7]: saved rax - saved by caller and bashed |
kvn@1938 | 975 | // * [tos + 8]: saved r10 (rscratch1) - saved by caller |
duke@435 | 976 | // * = popped on exit |
duke@435 | 977 | address generate_verify_oop() { |
duke@435 | 978 | StubCodeMark mark(this, "StubRoutines", "verify_oop"); |
duke@435 | 979 | address start = __ pc(); |
duke@435 | 980 | |
duke@435 | 981 | Label exit, error; |
duke@435 | 982 | |
never@739 | 983 | __ pushf(); |
duke@435 | 984 | __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr())); |
duke@435 | 985 | |
never@739 | 986 | __ push(r12); |
kvn@559 | 987 | |
duke@435 | 988 | // save c_rarg2 and c_rarg3 |
never@739 | 989 | __ push(c_rarg2); |
never@739 | 990 | __ push(c_rarg3); |
duke@435 | 991 | |
kvn@559 | 992 | enum { |
kvn@559 | 993 | // After previous pushes. |
kvn@559 | 994 | oop_to_verify = 6 * wordSize, |
kvn@559 | 995 | saved_rax = 7 * wordSize, |
kvn@1938 | 996 | saved_r10 = 8 * wordSize, |
kvn@559 | 997 | |
kvn@559 | 998 | // Before the call to MacroAssembler::debug(), see below. |
kvn@559 | 999 | return_addr = 16 * wordSize, |
kvn@559 | 1000 | error_msg = 17 * wordSize |
kvn@559 | 1001 | }; |
kvn@559 | 1002 | |
duke@435 | 1003 | // get object |
never@739 | 1004 | __ movptr(rax, Address(rsp, oop_to_verify)); |
duke@435 | 1005 | |
duke@435 | 1006 | // make sure object is 'reasonable' |
never@739 | 1007 | __ testptr(rax, rax); |
duke@435 | 1008 | __ jcc(Assembler::zero, exit); // if obj is NULL it is OK |
duke@435 | 1009 | // Check if the oop is in the right area of memory |
never@739 | 1010 | __ movptr(c_rarg2, rax); |
xlu@947 | 1011 | __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask()); |
never@739 | 1012 | __ andptr(c_rarg2, c_rarg3); |
xlu@947 | 1013 | __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits()); |
never@739 | 1014 | __ cmpptr(c_rarg2, c_rarg3); |
duke@435 | 1015 | __ jcc(Assembler::notZero, error); |
duke@435 | 1016 | |
kvn@559 | 1017 | // set r12 to heapbase for load_klass() |
kvn@559 | 1018 | __ reinit_heapbase(); |
kvn@559 | 1019 | |
coleenp@4037 | 1020 | // make sure klass is 'reasonable', which is not zero. |
coleenp@548 | 1021 | __ load_klass(rax, rax); // get klass |
never@739 | 1022 | __ testptr(rax, rax); |
duke@435 | 1023 | __ jcc(Assembler::zero, error); // if klass is NULL it is broken |
duke@435 | 1024 | |
duke@435 | 1025 | // return if everything seems ok |
duke@435 | 1026 | __ bind(exit); |
never@739 | 1027 | __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back |
kvn@1938 | 1028 | __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back |
never@739 | 1029 | __ pop(c_rarg3); // restore c_rarg3 |
never@739 | 1030 | __ pop(c_rarg2); // restore c_rarg2 |
never@739 | 1031 | __ pop(r12); // restore r12 |
never@739 | 1032 | __ popf(); // restore flags |
kvn@1938 | 1033 | __ ret(4 * wordSize); // pop caller saved stuff |
duke@435 | 1034 | |
duke@435 | 1035 | // handle errors |
duke@435 | 1036 | __ bind(error); |
never@739 | 1037 | __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back |
kvn@1938 | 1038 | __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back |
never@739 | 1039 | __ pop(c_rarg3); // get saved c_rarg3 back |
never@739 | 1040 | __ pop(c_rarg2); // get saved c_rarg2 back |
never@739 | 1041 | __ pop(r12); // get saved r12 back |
never@739 | 1042 | __ popf(); // get saved flags off stack -- |
duke@435 | 1043 | // will be ignored |
duke@435 | 1044 | |
never@739 | 1045 | __ pusha(); // push registers |
duke@435 | 1046 | // (rip is already |
duke@435 | 1047 | // already pushed) |
kvn@559 | 1048 | // debug(char* msg, int64_t pc, int64_t regs[]) |
duke@435 | 1049 | // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and |
duke@435 | 1050 | // pushed all the registers, so now the stack looks like: |
duke@435 | 1051 | // [tos + 0] 16 saved registers |
duke@435 | 1052 | // [tos + 16] return address |
kvn@559 | 1053 | // * [tos + 17] error message (char*) |
kvn@559 | 1054 | // * [tos + 18] object to verify (oop) |
kvn@559 | 1055 | // * [tos + 19] saved rax - saved by caller and bashed |
kvn@1938 | 1056 | // * [tos + 20] saved r10 (rscratch1) - saved by caller |
kvn@559 | 1057 | // * = popped on exit |
kvn@559 | 1058 | |
never@739 | 1059 | __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message |
never@739 | 1060 | __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address |
never@739 | 1061 | __ movq(c_rarg2, rsp); // pass address of regs on stack |
never@739 | 1062 | __ mov(r12, rsp); // remember rsp |
never@739 | 1063 | __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows |
never@739 | 1064 | __ andptr(rsp, -16); // align stack as required by ABI |
duke@435 | 1065 | BLOCK_COMMENT("call MacroAssembler::debug"); |
never@739 | 1066 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64))); |
never@739 | 1067 | __ mov(rsp, r12); // restore rsp |
never@739 | 1068 | __ popa(); // pop registers (includes r12) |
kvn@1938 | 1069 | __ ret(4 * wordSize); // pop caller saved stuff |
duke@435 | 1070 | |
duke@435 | 1071 | return start; |
duke@435 | 1072 | } |
duke@435 | 1073 | |
duke@435 | 1074 | // |
duke@435 | 1075 | // Verify that a register contains clean 32-bits positive value |
duke@435 | 1076 | // (high 32-bits are 0) so it could be used in 64-bits shifts. |
duke@435 | 1077 | // |
duke@435 | 1078 | // Input: |
duke@435 | 1079 | // Rint - 32-bits value |
duke@435 | 1080 | // Rtmp - scratch |
duke@435 | 1081 | // |
duke@435 | 1082 | void assert_clean_int(Register Rint, Register Rtmp) { |
duke@435 | 1083 | #ifdef ASSERT |
duke@435 | 1084 | Label L; |
duke@435 | 1085 | assert_different_registers(Rtmp, Rint); |
duke@435 | 1086 | __ movslq(Rtmp, Rint); |
duke@435 | 1087 | __ cmpq(Rtmp, Rint); |
kvn@559 | 1088 | __ jcc(Assembler::equal, L); |
duke@435 | 1089 | __ stop("high 32-bits of int value are not 0"); |
duke@435 | 1090 | __ bind(L); |
duke@435 | 1091 | #endif |
duke@435 | 1092 | } |
duke@435 | 1093 | |
duke@435 | 1094 | // Generate overlap test for array copy stubs |
duke@435 | 1095 | // |
duke@435 | 1096 | // Input: |
duke@435 | 1097 | // c_rarg0 - from |
duke@435 | 1098 | // c_rarg1 - to |
duke@435 | 1099 | // c_rarg2 - element count |
duke@435 | 1100 | // |
duke@435 | 1101 | // Output: |
duke@435 | 1102 | // rax - &from[element count - 1] |
duke@435 | 1103 | // |
duke@435 | 1104 | void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) { |
duke@435 | 1105 | assert(no_overlap_target != NULL, "must be generated"); |
duke@435 | 1106 | array_overlap_test(no_overlap_target, NULL, sf); |
duke@435 | 1107 | } |
duke@435 | 1108 | void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) { |
duke@435 | 1109 | array_overlap_test(NULL, &L_no_overlap, sf); |
duke@435 | 1110 | } |
duke@435 | 1111 | void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) { |
duke@435 | 1112 | const Register from = c_rarg0; |
duke@435 | 1113 | const Register to = c_rarg1; |
duke@435 | 1114 | const Register count = c_rarg2; |
duke@435 | 1115 | const Register end_from = rax; |
duke@435 | 1116 | |
never@739 | 1117 | __ cmpptr(to, from); |
never@739 | 1118 | __ lea(end_from, Address(from, count, sf, 0)); |
duke@435 | 1119 | if (NOLp == NULL) { |
duke@435 | 1120 | ExternalAddress no_overlap(no_overlap_target); |
duke@435 | 1121 | __ jump_cc(Assembler::belowEqual, no_overlap); |
never@739 | 1122 | __ cmpptr(to, end_from); |
duke@435 | 1123 | __ jump_cc(Assembler::aboveEqual, no_overlap); |
duke@435 | 1124 | } else { |
duke@435 | 1125 | __ jcc(Assembler::belowEqual, (*NOLp)); |
never@739 | 1126 | __ cmpptr(to, end_from); |
duke@435 | 1127 | __ jcc(Assembler::aboveEqual, (*NOLp)); |
duke@435 | 1128 | } |
duke@435 | 1129 | } |
duke@435 | 1130 | |
duke@435 | 1131 | // Shuffle first three arg regs on Windows into Linux/Solaris locations. |
duke@435 | 1132 | // |
duke@435 | 1133 | // Outputs: |
duke@435 | 1134 | // rdi - rcx |
duke@435 | 1135 | // rsi - rdx |
duke@435 | 1136 | // rdx - r8 |
duke@435 | 1137 | // rcx - r9 |
duke@435 | 1138 | // |
duke@435 | 1139 | // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter |
duke@435 | 1140 | // are non-volatile. r9 and r10 should not be used by the caller. |
duke@435 | 1141 | // |
duke@435 | 1142 | void setup_arg_regs(int nargs = 3) { |
duke@435 | 1143 | const Register saved_rdi = r9; |
duke@435 | 1144 | const Register saved_rsi = r10; |
duke@435 | 1145 | assert(nargs == 3 || nargs == 4, "else fix"); |
duke@435 | 1146 | #ifdef _WIN64 |
duke@435 | 1147 | assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9, |
duke@435 | 1148 | "unexpected argument registers"); |
duke@435 | 1149 | if (nargs >= 4) |
never@739 | 1150 | __ mov(rax, r9); // r9 is also saved_rdi |
never@739 | 1151 | __ movptr(saved_rdi, rdi); |
never@739 | 1152 | __ movptr(saved_rsi, rsi); |
never@739 | 1153 | __ mov(rdi, rcx); // c_rarg0 |
never@739 | 1154 | __ mov(rsi, rdx); // c_rarg1 |
never@739 | 1155 | __ mov(rdx, r8); // c_rarg2 |
duke@435 | 1156 | if (nargs >= 4) |
never@739 | 1157 | __ mov(rcx, rax); // c_rarg3 (via rax) |
duke@435 | 1158 | #else |
duke@435 | 1159 | assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx, |
duke@435 | 1160 | "unexpected argument registers"); |
duke@435 | 1161 | #endif |
duke@435 | 1162 | } |
duke@435 | 1163 | |
duke@435 | 1164 | void restore_arg_regs() { |
duke@435 | 1165 | const Register saved_rdi = r9; |
duke@435 | 1166 | const Register saved_rsi = r10; |
duke@435 | 1167 | #ifdef _WIN64 |
never@739 | 1168 | __ movptr(rdi, saved_rdi); |
never@739 | 1169 | __ movptr(rsi, saved_rsi); |
duke@435 | 1170 | #endif |
duke@435 | 1171 | } |
duke@435 | 1172 | |
duke@435 | 1173 | // Generate code for an array write pre barrier |
duke@435 | 1174 | // |
duke@435 | 1175 | // addr - starting address |
iveresov@2606 | 1176 | // count - element count |
iveresov@2606 | 1177 | // tmp - scratch register |
duke@435 | 1178 | // |
duke@435 | 1179 | // Destroy no registers! |
duke@435 | 1180 | // |
iveresov@2606 | 1181 | void gen_write_ref_array_pre_barrier(Register addr, Register count, bool dest_uninitialized) { |
duke@435 | 1182 | BarrierSet* bs = Universe::heap()->barrier_set(); |
duke@435 | 1183 | switch (bs->kind()) { |
duke@435 | 1184 | case BarrierSet::G1SATBCT: |
duke@435 | 1185 | case BarrierSet::G1SATBCTLogging: |
iveresov@2606 | 1186 | // With G1, don't generate the call if we statically know that the target in uninitialized |
iveresov@2606 | 1187 | if (!dest_uninitialized) { |
iveresov@2606 | 1188 | __ pusha(); // push registers |
iveresov@2606 | 1189 | if (count == c_rarg0) { |
iveresov@2606 | 1190 | if (addr == c_rarg1) { |
iveresov@2606 | 1191 | // exactly backwards!! |
iveresov@2606 | 1192 | __ xchgptr(c_rarg1, c_rarg0); |
iveresov@2606 | 1193 | } else { |
iveresov@2606 | 1194 | __ movptr(c_rarg1, count); |
iveresov@2606 | 1195 | __ movptr(c_rarg0, addr); |
iveresov@2606 | 1196 | } |
iveresov@2606 | 1197 | } else { |
iveresov@2606 | 1198 | __ movptr(c_rarg0, addr); |
iveresov@2606 | 1199 | __ movptr(c_rarg1, count); |
iveresov@2606 | 1200 | } |
iveresov@2606 | 1201 | __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2); |
iveresov@2606 | 1202 | __ popa(); |
duke@435 | 1203 | } |
iveresov@2606 | 1204 | break; |
duke@435 | 1205 | case BarrierSet::CardTableModRef: |
duke@435 | 1206 | case BarrierSet::CardTableExtension: |
duke@435 | 1207 | case BarrierSet::ModRef: |
duke@435 | 1208 | break; |
ysr@777 | 1209 | default: |
duke@435 | 1210 | ShouldNotReachHere(); |
duke@435 | 1211 | |
duke@435 | 1212 | } |
duke@435 | 1213 | } |
duke@435 | 1214 | |
duke@435 | 1215 | // |
duke@435 | 1216 | // Generate code for an array write post barrier |
duke@435 | 1217 | // |
duke@435 | 1218 | // Input: |
duke@435 | 1219 | // start - register containing starting address of destination array |
kvn@5156 | 1220 | // count - elements count |
duke@435 | 1221 | // scratch - scratch register |
duke@435 | 1222 | // |
duke@435 | 1223 | // The input registers are overwritten. |
kvn@5156 | 1224 | // |
kvn@5156 | 1225 | void gen_write_ref_array_post_barrier(Register start, Register count, Register scratch) { |
kvn@5156 | 1226 | assert_different_registers(start, count, scratch); |
duke@435 | 1227 | BarrierSet* bs = Universe::heap()->barrier_set(); |
duke@435 | 1228 | switch (bs->kind()) { |
duke@435 | 1229 | case BarrierSet::G1SATBCT: |
duke@435 | 1230 | case BarrierSet::G1SATBCTLogging: |
duke@435 | 1231 | { |
kvn@5156 | 1232 | __ pusha(); // push registers (overkill) |
kvn@5156 | 1233 | if (c_rarg0 == count) { // On win64 c_rarg0 == rcx |
kvn@5156 | 1234 | assert_different_registers(c_rarg1, start); |
kvn@5156 | 1235 | __ mov(c_rarg1, count); |
kvn@5156 | 1236 | __ mov(c_rarg0, start); |
kvn@5156 | 1237 | } else { |
kvn@5156 | 1238 | assert_different_registers(c_rarg0, count); |
kvn@5156 | 1239 | __ mov(c_rarg0, start); |
kvn@5156 | 1240 | __ mov(c_rarg1, count); |
kvn@5156 | 1241 | } |
apetrusenko@1627 | 1242 | __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2); |
never@739 | 1243 | __ popa(); |
duke@435 | 1244 | } |
duke@435 | 1245 | break; |
duke@435 | 1246 | case BarrierSet::CardTableModRef: |
duke@435 | 1247 | case BarrierSet::CardTableExtension: |
duke@435 | 1248 | { |
duke@435 | 1249 | CardTableModRefBS* ct = (CardTableModRefBS*)bs; |
duke@435 | 1250 | assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); |
duke@435 | 1251 | |
duke@435 | 1252 | Label L_loop; |
kvn@5156 | 1253 | const Register end = count; |
kvn@5156 | 1254 | |
kvn@5156 | 1255 | __ leaq(end, Address(start, count, TIMES_OOP, 0)); // end == start+count*oop_size |
kvn@5156 | 1256 | __ subptr(end, BytesPerHeapOop); // end - 1 to make inclusive |
kvn@5156 | 1257 | __ shrptr(start, CardTableModRefBS::card_shift); |
kvn@5156 | 1258 | __ shrptr(end, CardTableModRefBS::card_shift); |
kvn@5156 | 1259 | __ subptr(end, start); // end --> cards count |
kvn@5156 | 1260 | |
kvn@5156 | 1261 | int64_t disp = (int64_t) ct->byte_map_base; |
kvn@5156 | 1262 | __ mov64(scratch, disp); |
never@739 | 1263 | __ addptr(start, scratch); |
duke@435 | 1264 | __ BIND(L_loop); |
duke@435 | 1265 | __ movb(Address(start, count, Address::times_1), 0); |
never@739 | 1266 | __ decrement(count); |
duke@435 | 1267 | __ jcc(Assembler::greaterEqual, L_loop); |
duke@435 | 1268 | } |
ysr@777 | 1269 | break; |
ysr@777 | 1270 | default: |
ysr@777 | 1271 | ShouldNotReachHere(); |
ysr@777 | 1272 | |
ysr@777 | 1273 | } |
ysr@777 | 1274 | } |
duke@435 | 1275 | |
kvn@840 | 1276 | |
duke@435 | 1277 | // Copy big chunks forward |
duke@435 | 1278 | // |
duke@435 | 1279 | // Inputs: |
duke@435 | 1280 | // end_from - source arrays end address |
duke@435 | 1281 | // end_to - destination array end address |
duke@435 | 1282 | // qword_count - 64-bits element count, negative |
duke@435 | 1283 | // to - scratch |
kvn@4411 | 1284 | // L_copy_bytes - entry label |
duke@435 | 1285 | // L_copy_8_bytes - exit label |
duke@435 | 1286 | // |
kvn@4411 | 1287 | void copy_bytes_forward(Register end_from, Register end_to, |
duke@435 | 1288 | Register qword_count, Register to, |
kvn@4411 | 1289 | Label& L_copy_bytes, Label& L_copy_8_bytes) { |
duke@435 | 1290 | DEBUG_ONLY(__ stop("enter at entry label, not here")); |
duke@435 | 1291 | Label L_loop; |
kvn@1800 | 1292 | __ align(OptoLoopAlignment); |
kvn@4411 | 1293 | if (UseUnalignedLoadStores) { |
kvn@4411 | 1294 | Label L_end; |
kvn@4411 | 1295 | // Copy 64-bytes per iteration |
kvn@4411 | 1296 | __ BIND(L_loop); |
kvn@4411 | 1297 | if (UseAVX >= 2) { |
kvn@4411 | 1298 | __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56)); |
kvn@4411 | 1299 | __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0); |
kvn@4411 | 1300 | __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24)); |
kvn@4411 | 1301 | __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1); |
kvn@4411 | 1302 | } else { |
kvn@4411 | 1303 | __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56)); |
kvn@4411 | 1304 | __ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0); |
kvn@4411 | 1305 | __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40)); |
kvn@4411 | 1306 | __ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1); |
kvn@4411 | 1307 | __ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24)); |
kvn@4411 | 1308 | __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2); |
kvn@4411 | 1309 | __ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8)); |
kvn@4411 | 1310 | __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3); |
kvn@4411 | 1311 | } |
kvn@4411 | 1312 | __ BIND(L_copy_bytes); |
kvn@4411 | 1313 | __ addptr(qword_count, 8); |
kvn@4411 | 1314 | __ jcc(Assembler::lessEqual, L_loop); |
kvn@4411 | 1315 | __ subptr(qword_count, 4); // sub(8) and add(4) |
kvn@4411 | 1316 | __ jccb(Assembler::greater, L_end); |
kvn@4411 | 1317 | // Copy trailing 32 bytes |
kvn@4411 | 1318 | if (UseAVX >= 2) { |
kvn@4411 | 1319 | __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24)); |
kvn@4411 | 1320 | __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0); |
kvn@4411 | 1321 | } else { |
kvn@4411 | 1322 | __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24)); |
kvn@4411 | 1323 | __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0); |
kvn@4411 | 1324 | __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8)); |
kvn@4411 | 1325 | __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1); |
kvn@4411 | 1326 | } |
kvn@4411 | 1327 | __ addptr(qword_count, 4); |
kvn@4411 | 1328 | __ BIND(L_end); |
kvn@4873 | 1329 | if (UseAVX >= 2) { |
kvn@4873 | 1330 | // clean upper bits of YMM registers |
kvn@4873 | 1331 | __ vzeroupper(); |
kvn@4873 | 1332 | } |
kvn@840 | 1333 | } else { |
kvn@4411 | 1334 | // Copy 32-bytes per iteration |
kvn@4411 | 1335 | __ BIND(L_loop); |
kvn@840 | 1336 | __ movq(to, Address(end_from, qword_count, Address::times_8, -24)); |
kvn@840 | 1337 | __ movq(Address(end_to, qword_count, Address::times_8, -24), to); |
kvn@840 | 1338 | __ movq(to, Address(end_from, qword_count, Address::times_8, -16)); |
kvn@840 | 1339 | __ movq(Address(end_to, qword_count, Address::times_8, -16), to); |
kvn@840 | 1340 | __ movq(to, Address(end_from, qword_count, Address::times_8, - 8)); |
kvn@840 | 1341 | __ movq(Address(end_to, qword_count, Address::times_8, - 8), to); |
kvn@840 | 1342 | __ movq(to, Address(end_from, qword_count, Address::times_8, - 0)); |
kvn@840 | 1343 | __ movq(Address(end_to, qword_count, Address::times_8, - 0), to); |
kvn@4411 | 1344 | |
kvn@4411 | 1345 | __ BIND(L_copy_bytes); |
kvn@4411 | 1346 | __ addptr(qword_count, 4); |
kvn@4411 | 1347 | __ jcc(Assembler::lessEqual, L_loop); |
kvn@840 | 1348 | } |
never@739 | 1349 | __ subptr(qword_count, 4); |
duke@435 | 1350 | __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords |
duke@435 | 1351 | } |
duke@435 | 1352 | |
duke@435 | 1353 | // Copy big chunks backward |
duke@435 | 1354 | // |
duke@435 | 1355 | // Inputs: |
duke@435 | 1356 | // from - source arrays address |
duke@435 | 1357 | // dest - destination array address |
duke@435 | 1358 | // qword_count - 64-bits element count |
duke@435 | 1359 | // to - scratch |
kvn@4411 | 1360 | // L_copy_bytes - entry label |
duke@435 | 1361 | // L_copy_8_bytes - exit label |
duke@435 | 1362 | // |
kvn@4411 | 1363 | void copy_bytes_backward(Register from, Register dest, |
duke@435 | 1364 | Register qword_count, Register to, |
kvn@4411 | 1365 | Label& L_copy_bytes, Label& L_copy_8_bytes) { |
duke@435 | 1366 | DEBUG_ONLY(__ stop("enter at entry label, not here")); |
duke@435 | 1367 | Label L_loop; |
kvn@1800 | 1368 | __ align(OptoLoopAlignment); |
kvn@4411 | 1369 | if (UseUnalignedLoadStores) { |
kvn@4411 | 1370 | Label L_end; |
kvn@4411 | 1371 | // Copy 64-bytes per iteration |
kvn@4411 | 1372 | __ BIND(L_loop); |
kvn@4411 | 1373 | if (UseAVX >= 2) { |
kvn@4411 | 1374 | __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32)); |
kvn@4411 | 1375 | __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0); |
kvn@4411 | 1376 | __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0)); |
kvn@4411 | 1377 | __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1); |
kvn@4411 | 1378 | } else { |
kvn@4411 | 1379 | __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48)); |
kvn@4411 | 1380 | __ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0); |
kvn@4411 | 1381 | __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32)); |
kvn@4411 | 1382 | __ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1); |
kvn@4411 | 1383 | __ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16)); |
kvn@4411 | 1384 | __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2); |
kvn@4411 | 1385 | __ movdqu(xmm3, Address(from, qword_count, Address::times_8, 0)); |
kvn@4411 | 1386 | __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm3); |
kvn@4411 | 1387 | } |
kvn@4411 | 1388 | __ BIND(L_copy_bytes); |
kvn@4411 | 1389 | __ subptr(qword_count, 8); |
kvn@4411 | 1390 | __ jcc(Assembler::greaterEqual, L_loop); |
kvn@4411 | 1391 | |
kvn@4411 | 1392 | __ addptr(qword_count, 4); // add(8) and sub(4) |
kvn@4411 | 1393 | __ jccb(Assembler::less, L_end); |
kvn@4411 | 1394 | // Copy trailing 32 bytes |
kvn@4411 | 1395 | if (UseAVX >= 2) { |
kvn@4411 | 1396 | __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0)); |
kvn@4411 | 1397 | __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0); |
kvn@4411 | 1398 | } else { |
kvn@4411 | 1399 | __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16)); |
kvn@4411 | 1400 | __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0); |
kvn@4411 | 1401 | __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0)); |
kvn@4411 | 1402 | __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1); |
kvn@4411 | 1403 | } |
kvn@4411 | 1404 | __ subptr(qword_count, 4); |
kvn@4411 | 1405 | __ BIND(L_end); |
kvn@4873 | 1406 | if (UseAVX >= 2) { |
kvn@4873 | 1407 | // clean upper bits of YMM registers |
kvn@4873 | 1408 | __ vzeroupper(); |
kvn@4873 | 1409 | } |
kvn@840 | 1410 | } else { |
kvn@4411 | 1411 | // Copy 32-bytes per iteration |
kvn@4411 | 1412 | __ BIND(L_loop); |
kvn@840 | 1413 | __ movq(to, Address(from, qword_count, Address::times_8, 24)); |
kvn@840 | 1414 | __ movq(Address(dest, qword_count, Address::times_8, 24), to); |
kvn@840 | 1415 | __ movq(to, Address(from, qword_count, Address::times_8, 16)); |
kvn@840 | 1416 | __ movq(Address(dest, qword_count, Address::times_8, 16), to); |
kvn@840 | 1417 | __ movq(to, Address(from, qword_count, Address::times_8, 8)); |
kvn@840 | 1418 | __ movq(Address(dest, qword_count, Address::times_8, 8), to); |
kvn@840 | 1419 | __ movq(to, Address(from, qword_count, Address::times_8, 0)); |
kvn@840 | 1420 | __ movq(Address(dest, qword_count, Address::times_8, 0), to); |
kvn@4411 | 1421 | |
kvn@4411 | 1422 | __ BIND(L_copy_bytes); |
kvn@4411 | 1423 | __ subptr(qword_count, 4); |
kvn@4411 | 1424 | __ jcc(Assembler::greaterEqual, L_loop); |
kvn@840 | 1425 | } |
never@739 | 1426 | __ addptr(qword_count, 4); |
duke@435 | 1427 | __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords |
duke@435 | 1428 | } |
duke@435 | 1429 | |
duke@435 | 1430 | |
duke@435 | 1431 | // Arguments: |
duke@435 | 1432 | // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary |
duke@435 | 1433 | // ignored |
duke@435 | 1434 | // name - stub name string |
duke@435 | 1435 | // |
duke@435 | 1436 | // Inputs: |
duke@435 | 1437 | // c_rarg0 - source array address |
duke@435 | 1438 | // c_rarg1 - destination array address |
duke@435 | 1439 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1440 | // |
duke@435 | 1441 | // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries, |
duke@435 | 1442 | // we let the hardware handle it. The one to eight bytes within words, |
duke@435 | 1443 | // dwords or qwords that span cache line boundaries will still be loaded |
duke@435 | 1444 | // and stored atomically. |
duke@435 | 1445 | // |
duke@435 | 1446 | // Side Effects: |
duke@435 | 1447 | // disjoint_byte_copy_entry is set to the no-overlap entry point |
duke@435 | 1448 | // used by generate_conjoint_byte_copy(). |
duke@435 | 1449 | // |
iveresov@2595 | 1450 | address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) { |
duke@435 | 1451 | __ align(CodeEntryAlignment); |
duke@435 | 1452 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1453 | address start = __ pc(); |
duke@435 | 1454 | |
kvn@4411 | 1455 | Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes; |
duke@435 | 1456 | Label L_copy_byte, L_exit; |
duke@435 | 1457 | const Register from = rdi; // source array address |
duke@435 | 1458 | const Register to = rsi; // destination array address |
duke@435 | 1459 | const Register count = rdx; // elements count |
duke@435 | 1460 | const Register byte_count = rcx; |
duke@435 | 1461 | const Register qword_count = count; |
duke@435 | 1462 | const Register end_from = from; // source array end address |
duke@435 | 1463 | const Register end_to = to; // destination array end address |
duke@435 | 1464 | // End pointers are inclusive, and if count is not zero they point |
duke@435 | 1465 | // to the last unit copied: end_to[0] := end_from[0] |
duke@435 | 1466 | |
duke@435 | 1467 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1468 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 1469 | |
iveresov@2595 | 1470 | if (entry != NULL) { |
iveresov@2595 | 1471 | *entry = __ pc(); |
iveresov@2595 | 1472 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1473 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1474 | } |
duke@435 | 1475 | |
duke@435 | 1476 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 1477 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 1478 | |
duke@435 | 1479 | // 'from', 'to' and 'count' are now valid |
never@739 | 1480 | __ movptr(byte_count, count); |
never@739 | 1481 | __ shrptr(count, 3); // count => qword_count |
duke@435 | 1482 | |
duke@435 | 1483 | // Copy from low to high addresses. Use 'to' as scratch. |
never@739 | 1484 | __ lea(end_from, Address(from, qword_count, Address::times_8, -8)); |
never@739 | 1485 | __ lea(end_to, Address(to, qword_count, Address::times_8, -8)); |
never@739 | 1486 | __ negptr(qword_count); // make the count negative |
kvn@4411 | 1487 | __ jmp(L_copy_bytes); |
duke@435 | 1488 | |
duke@435 | 1489 | // Copy trailing qwords |
duke@435 | 1490 | __ BIND(L_copy_8_bytes); |
duke@435 | 1491 | __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); |
duke@435 | 1492 | __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); |
never@739 | 1493 | __ increment(qword_count); |
duke@435 | 1494 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 1495 | |
duke@435 | 1496 | // Check for and copy trailing dword |
duke@435 | 1497 | __ BIND(L_copy_4_bytes); |
never@739 | 1498 | __ testl(byte_count, 4); |
duke@435 | 1499 | __ jccb(Assembler::zero, L_copy_2_bytes); |
duke@435 | 1500 | __ movl(rax, Address(end_from, 8)); |
duke@435 | 1501 | __ movl(Address(end_to, 8), rax); |
duke@435 | 1502 | |
never@739 | 1503 | __ addptr(end_from, 4); |
never@739 | 1504 | __ addptr(end_to, 4); |
duke@435 | 1505 | |
duke@435 | 1506 | // Check for and copy trailing word |
duke@435 | 1507 | __ BIND(L_copy_2_bytes); |
never@739 | 1508 | __ testl(byte_count, 2); |
duke@435 | 1509 | __ jccb(Assembler::zero, L_copy_byte); |
duke@435 | 1510 | __ movw(rax, Address(end_from, 8)); |
duke@435 | 1511 | __ movw(Address(end_to, 8), rax); |
duke@435 | 1512 | |
never@739 | 1513 | __ addptr(end_from, 2); |
never@739 | 1514 | __ addptr(end_to, 2); |
duke@435 | 1515 | |
duke@435 | 1516 | // Check for and copy trailing byte |
duke@435 | 1517 | __ BIND(L_copy_byte); |
never@739 | 1518 | __ testl(byte_count, 1); |
duke@435 | 1519 | __ jccb(Assembler::zero, L_exit); |
duke@435 | 1520 | __ movb(rax, Address(end_from, 8)); |
duke@435 | 1521 | __ movb(Address(end_to, 8), rax); |
duke@435 | 1522 | |
duke@435 | 1523 | __ BIND(L_exit); |
duke@435 | 1524 | restore_arg_regs(); |
never@3314 | 1525 | inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1526 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1527 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1528 | __ ret(0); |
duke@435 | 1529 | |
kvn@4411 | 1530 | // Copy in multi-bytes chunks |
kvn@4411 | 1531 | copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); |
duke@435 | 1532 | __ jmp(L_copy_4_bytes); |
duke@435 | 1533 | |
duke@435 | 1534 | return start; |
duke@435 | 1535 | } |
duke@435 | 1536 | |
duke@435 | 1537 | // Arguments: |
duke@435 | 1538 | // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary |
duke@435 | 1539 | // ignored |
duke@435 | 1540 | // name - stub name string |
duke@435 | 1541 | // |
duke@435 | 1542 | // Inputs: |
duke@435 | 1543 | // c_rarg0 - source array address |
duke@435 | 1544 | // c_rarg1 - destination array address |
duke@435 | 1545 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1546 | // |
duke@435 | 1547 | // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries, |
duke@435 | 1548 | // we let the hardware handle it. The one to eight bytes within words, |
duke@435 | 1549 | // dwords or qwords that span cache line boundaries will still be loaded |
duke@435 | 1550 | // and stored atomically. |
duke@435 | 1551 | // |
iveresov@2595 | 1552 | address generate_conjoint_byte_copy(bool aligned, address nooverlap_target, |
iveresov@2595 | 1553 | address* entry, const char *name) { |
duke@435 | 1554 | __ align(CodeEntryAlignment); |
duke@435 | 1555 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1556 | address start = __ pc(); |
duke@435 | 1557 | |
kvn@4411 | 1558 | Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes; |
duke@435 | 1559 | const Register from = rdi; // source array address |
duke@435 | 1560 | const Register to = rsi; // destination array address |
duke@435 | 1561 | const Register count = rdx; // elements count |
duke@435 | 1562 | const Register byte_count = rcx; |
duke@435 | 1563 | const Register qword_count = count; |
duke@435 | 1564 | |
duke@435 | 1565 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1566 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 1567 | |
iveresov@2595 | 1568 | if (entry != NULL) { |
iveresov@2595 | 1569 | *entry = __ pc(); |
iveresov@2595 | 1570 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1571 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1572 | } |
iveresov@2595 | 1573 | |
iveresov@2595 | 1574 | array_overlap_test(nooverlap_target, Address::times_1); |
duke@435 | 1575 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 1576 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 1577 | |
duke@435 | 1578 | // 'from', 'to' and 'count' are now valid |
never@739 | 1579 | __ movptr(byte_count, count); |
never@739 | 1580 | __ shrptr(count, 3); // count => qword_count |
duke@435 | 1581 | |
duke@435 | 1582 | // Copy from high to low addresses. |
duke@435 | 1583 | |
duke@435 | 1584 | // Check for and copy trailing byte |
never@739 | 1585 | __ testl(byte_count, 1); |
duke@435 | 1586 | __ jcc(Assembler::zero, L_copy_2_bytes); |
duke@435 | 1587 | __ movb(rax, Address(from, byte_count, Address::times_1, -1)); |
duke@435 | 1588 | __ movb(Address(to, byte_count, Address::times_1, -1), rax); |
never@739 | 1589 | __ decrement(byte_count); // Adjust for possible trailing word |
duke@435 | 1590 | |
duke@435 | 1591 | // Check for and copy trailing word |
duke@435 | 1592 | __ BIND(L_copy_2_bytes); |
never@739 | 1593 | __ testl(byte_count, 2); |
duke@435 | 1594 | __ jcc(Assembler::zero, L_copy_4_bytes); |
duke@435 | 1595 | __ movw(rax, Address(from, byte_count, Address::times_1, -2)); |
duke@435 | 1596 | __ movw(Address(to, byte_count, Address::times_1, -2), rax); |
duke@435 | 1597 | |
duke@435 | 1598 | // Check for and copy trailing dword |
duke@435 | 1599 | __ BIND(L_copy_4_bytes); |
never@739 | 1600 | __ testl(byte_count, 4); |
kvn@4411 | 1601 | __ jcc(Assembler::zero, L_copy_bytes); |
duke@435 | 1602 | __ movl(rax, Address(from, qword_count, Address::times_8)); |
duke@435 | 1603 | __ movl(Address(to, qword_count, Address::times_8), rax); |
kvn@4411 | 1604 | __ jmp(L_copy_bytes); |
duke@435 | 1605 | |
duke@435 | 1606 | // Copy trailing qwords |
duke@435 | 1607 | __ BIND(L_copy_8_bytes); |
duke@435 | 1608 | __ movq(rax, Address(from, qword_count, Address::times_8, -8)); |
duke@435 | 1609 | __ movq(Address(to, qword_count, Address::times_8, -8), rax); |
never@739 | 1610 | __ decrement(qword_count); |
duke@435 | 1611 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 1612 | |
duke@435 | 1613 | restore_arg_regs(); |
never@3314 | 1614 | inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1615 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1616 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1617 | __ ret(0); |
duke@435 | 1618 | |
kvn@4411 | 1619 | // Copy in multi-bytes chunks |
kvn@4411 | 1620 | copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); |
duke@435 | 1621 | |
duke@435 | 1622 | restore_arg_regs(); |
never@3314 | 1623 | inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1624 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1625 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1626 | __ ret(0); |
duke@435 | 1627 | |
duke@435 | 1628 | return start; |
duke@435 | 1629 | } |
duke@435 | 1630 | |
duke@435 | 1631 | // Arguments: |
duke@435 | 1632 | // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary |
duke@435 | 1633 | // ignored |
duke@435 | 1634 | // name - stub name string |
duke@435 | 1635 | // |
duke@435 | 1636 | // Inputs: |
duke@435 | 1637 | // c_rarg0 - source array address |
duke@435 | 1638 | // c_rarg1 - destination array address |
duke@435 | 1639 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1640 | // |
duke@435 | 1641 | // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we |
duke@435 | 1642 | // let the hardware handle it. The two or four words within dwords |
duke@435 | 1643 | // or qwords that span cache line boundaries will still be loaded |
duke@435 | 1644 | // and stored atomically. |
duke@435 | 1645 | // |
duke@435 | 1646 | // Side Effects: |
duke@435 | 1647 | // disjoint_short_copy_entry is set to the no-overlap entry point |
duke@435 | 1648 | // used by generate_conjoint_short_copy(). |
duke@435 | 1649 | // |
iveresov@2595 | 1650 | address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) { |
duke@435 | 1651 | __ align(CodeEntryAlignment); |
duke@435 | 1652 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1653 | address start = __ pc(); |
duke@435 | 1654 | |
kvn@4411 | 1655 | Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit; |
duke@435 | 1656 | const Register from = rdi; // source array address |
duke@435 | 1657 | const Register to = rsi; // destination array address |
duke@435 | 1658 | const Register count = rdx; // elements count |
duke@435 | 1659 | const Register word_count = rcx; |
duke@435 | 1660 | const Register qword_count = count; |
duke@435 | 1661 | const Register end_from = from; // source array end address |
duke@435 | 1662 | const Register end_to = to; // destination array end address |
duke@435 | 1663 | // End pointers are inclusive, and if count is not zero they point |
duke@435 | 1664 | // to the last unit copied: end_to[0] := end_from[0] |
duke@435 | 1665 | |
duke@435 | 1666 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1667 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 1668 | |
iveresov@2595 | 1669 | if (entry != NULL) { |
iveresov@2595 | 1670 | *entry = __ pc(); |
iveresov@2595 | 1671 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1672 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1673 | } |
duke@435 | 1674 | |
duke@435 | 1675 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 1676 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 1677 | |
duke@435 | 1678 | // 'from', 'to' and 'count' are now valid |
never@739 | 1679 | __ movptr(word_count, count); |
never@739 | 1680 | __ shrptr(count, 2); // count => qword_count |
duke@435 | 1681 | |
duke@435 | 1682 | // Copy from low to high addresses. Use 'to' as scratch. |
never@739 | 1683 | __ lea(end_from, Address(from, qword_count, Address::times_8, -8)); |
never@739 | 1684 | __ lea(end_to, Address(to, qword_count, Address::times_8, -8)); |
never@739 | 1685 | __ negptr(qword_count); |
kvn@4411 | 1686 | __ jmp(L_copy_bytes); |
duke@435 | 1687 | |
duke@435 | 1688 | // Copy trailing qwords |
duke@435 | 1689 | __ BIND(L_copy_8_bytes); |
duke@435 | 1690 | __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); |
duke@435 | 1691 | __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); |
never@739 | 1692 | __ increment(qword_count); |
duke@435 | 1693 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 1694 | |
duke@435 | 1695 | // Original 'dest' is trashed, so we can't use it as a |
duke@435 | 1696 | // base register for a possible trailing word copy |
duke@435 | 1697 | |
duke@435 | 1698 | // Check for and copy trailing dword |
duke@435 | 1699 | __ BIND(L_copy_4_bytes); |
never@739 | 1700 | __ testl(word_count, 2); |
duke@435 | 1701 | __ jccb(Assembler::zero, L_copy_2_bytes); |
duke@435 | 1702 | __ movl(rax, Address(end_from, 8)); |
duke@435 | 1703 | __ movl(Address(end_to, 8), rax); |
duke@435 | 1704 | |
never@739 | 1705 | __ addptr(end_from, 4); |
never@739 | 1706 | __ addptr(end_to, 4); |
duke@435 | 1707 | |
duke@435 | 1708 | // Check for and copy trailing word |
duke@435 | 1709 | __ BIND(L_copy_2_bytes); |
never@739 | 1710 | __ testl(word_count, 1); |
duke@435 | 1711 | __ jccb(Assembler::zero, L_exit); |
duke@435 | 1712 | __ movw(rax, Address(end_from, 8)); |
duke@435 | 1713 | __ movw(Address(end_to, 8), rax); |
duke@435 | 1714 | |
duke@435 | 1715 | __ BIND(L_exit); |
duke@435 | 1716 | restore_arg_regs(); |
never@3314 | 1717 | inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1718 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1719 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1720 | __ ret(0); |
duke@435 | 1721 | |
kvn@4411 | 1722 | // Copy in multi-bytes chunks |
kvn@4411 | 1723 | copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); |
duke@435 | 1724 | __ jmp(L_copy_4_bytes); |
duke@435 | 1725 | |
duke@435 | 1726 | return start; |
duke@435 | 1727 | } |
duke@435 | 1728 | |
never@2118 | 1729 | address generate_fill(BasicType t, bool aligned, const char *name) { |
never@2118 | 1730 | __ align(CodeEntryAlignment); |
never@2118 | 1731 | StubCodeMark mark(this, "StubRoutines", name); |
never@2118 | 1732 | address start = __ pc(); |
never@2118 | 1733 | |
never@2118 | 1734 | BLOCK_COMMENT("Entry:"); |
never@2118 | 1735 | |
never@2118 | 1736 | const Register to = c_rarg0; // source array address |
never@2118 | 1737 | const Register value = c_rarg1; // value |
never@2118 | 1738 | const Register count = c_rarg2; // elements count |
never@2118 | 1739 | |
never@2118 | 1740 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
never@2118 | 1741 | |
never@2118 | 1742 | __ generate_fill(t, aligned, to, value, count, rax, xmm0); |
never@2118 | 1743 | |
never@2118 | 1744 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
never@2118 | 1745 | __ ret(0); |
never@2118 | 1746 | return start; |
never@2118 | 1747 | } |
never@2118 | 1748 | |
duke@435 | 1749 | // Arguments: |
duke@435 | 1750 | // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary |
duke@435 | 1751 | // ignored |
duke@435 | 1752 | // name - stub name string |
duke@435 | 1753 | // |
duke@435 | 1754 | // Inputs: |
duke@435 | 1755 | // c_rarg0 - source array address |
duke@435 | 1756 | // c_rarg1 - destination array address |
duke@435 | 1757 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1758 | // |
duke@435 | 1759 | // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we |
duke@435 | 1760 | // let the hardware handle it. The two or four words within dwords |
duke@435 | 1761 | // or qwords that span cache line boundaries will still be loaded |
duke@435 | 1762 | // and stored atomically. |
duke@435 | 1763 | // |
iveresov@2595 | 1764 | address generate_conjoint_short_copy(bool aligned, address nooverlap_target, |
iveresov@2595 | 1765 | address *entry, const char *name) { |
duke@435 | 1766 | __ align(CodeEntryAlignment); |
duke@435 | 1767 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1768 | address start = __ pc(); |
duke@435 | 1769 | |
kvn@4411 | 1770 | Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes; |
duke@435 | 1771 | const Register from = rdi; // source array address |
duke@435 | 1772 | const Register to = rsi; // destination array address |
duke@435 | 1773 | const Register count = rdx; // elements count |
duke@435 | 1774 | const Register word_count = rcx; |
duke@435 | 1775 | const Register qword_count = count; |
duke@435 | 1776 | |
duke@435 | 1777 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1778 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 1779 | |
iveresov@2595 | 1780 | if (entry != NULL) { |
iveresov@2595 | 1781 | *entry = __ pc(); |
iveresov@2595 | 1782 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1783 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1784 | } |
iveresov@2595 | 1785 | |
iveresov@2595 | 1786 | array_overlap_test(nooverlap_target, Address::times_2); |
duke@435 | 1787 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 1788 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 1789 | |
duke@435 | 1790 | // 'from', 'to' and 'count' are now valid |
never@739 | 1791 | __ movptr(word_count, count); |
never@739 | 1792 | __ shrptr(count, 2); // count => qword_count |
duke@435 | 1793 | |
duke@435 | 1794 | // Copy from high to low addresses. Use 'to' as scratch. |
duke@435 | 1795 | |
duke@435 | 1796 | // Check for and copy trailing word |
never@739 | 1797 | __ testl(word_count, 1); |
duke@435 | 1798 | __ jccb(Assembler::zero, L_copy_4_bytes); |
duke@435 | 1799 | __ movw(rax, Address(from, word_count, Address::times_2, -2)); |
duke@435 | 1800 | __ movw(Address(to, word_count, Address::times_2, -2), rax); |
duke@435 | 1801 | |
duke@435 | 1802 | // Check for and copy trailing dword |
duke@435 | 1803 | __ BIND(L_copy_4_bytes); |
never@739 | 1804 | __ testl(word_count, 2); |
kvn@4411 | 1805 | __ jcc(Assembler::zero, L_copy_bytes); |
duke@435 | 1806 | __ movl(rax, Address(from, qword_count, Address::times_8)); |
duke@435 | 1807 | __ movl(Address(to, qword_count, Address::times_8), rax); |
kvn@4411 | 1808 | __ jmp(L_copy_bytes); |
duke@435 | 1809 | |
duke@435 | 1810 | // Copy trailing qwords |
duke@435 | 1811 | __ BIND(L_copy_8_bytes); |
duke@435 | 1812 | __ movq(rax, Address(from, qword_count, Address::times_8, -8)); |
duke@435 | 1813 | __ movq(Address(to, qword_count, Address::times_8, -8), rax); |
never@739 | 1814 | __ decrement(qword_count); |
duke@435 | 1815 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 1816 | |
duke@435 | 1817 | restore_arg_regs(); |
never@3314 | 1818 | inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1819 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1820 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1821 | __ ret(0); |
duke@435 | 1822 | |
kvn@4411 | 1823 | // Copy in multi-bytes chunks |
kvn@4411 | 1824 | copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); |
duke@435 | 1825 | |
duke@435 | 1826 | restore_arg_regs(); |
never@3314 | 1827 | inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1828 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1829 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1830 | __ ret(0); |
duke@435 | 1831 | |
duke@435 | 1832 | return start; |
duke@435 | 1833 | } |
duke@435 | 1834 | |
duke@435 | 1835 | // Arguments: |
duke@435 | 1836 | // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary |
duke@435 | 1837 | // ignored |
coleenp@548 | 1838 | // is_oop - true => oop array, so generate store check code |
duke@435 | 1839 | // name - stub name string |
duke@435 | 1840 | // |
duke@435 | 1841 | // Inputs: |
duke@435 | 1842 | // c_rarg0 - source array address |
duke@435 | 1843 | // c_rarg1 - destination array address |
duke@435 | 1844 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1845 | // |
duke@435 | 1846 | // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let |
duke@435 | 1847 | // the hardware handle it. The two dwords within qwords that span |
duke@435 | 1848 | // cache line boundaries will still be loaded and stored atomicly. |
duke@435 | 1849 | // |
duke@435 | 1850 | // Side Effects: |
duke@435 | 1851 | // disjoint_int_copy_entry is set to the no-overlap entry point |
coleenp@548 | 1852 | // used by generate_conjoint_int_oop_copy(). |
duke@435 | 1853 | // |
iveresov@2606 | 1854 | address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry, |
iveresov@2606 | 1855 | const char *name, bool dest_uninitialized = false) { |
duke@435 | 1856 | __ align(CodeEntryAlignment); |
duke@435 | 1857 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1858 | address start = __ pc(); |
duke@435 | 1859 | |
kvn@4411 | 1860 | Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit; |
duke@435 | 1861 | const Register from = rdi; // source array address |
duke@435 | 1862 | const Register to = rsi; // destination array address |
duke@435 | 1863 | const Register count = rdx; // elements count |
duke@435 | 1864 | const Register dword_count = rcx; |
duke@435 | 1865 | const Register qword_count = count; |
duke@435 | 1866 | const Register end_from = from; // source array end address |
duke@435 | 1867 | const Register end_to = to; // destination array end address |
coleenp@548 | 1868 | const Register saved_to = r11; // saved destination array address |
duke@435 | 1869 | // End pointers are inclusive, and if count is not zero they point |
duke@435 | 1870 | // to the last unit copied: end_to[0] := end_from[0] |
duke@435 | 1871 | |
duke@435 | 1872 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1873 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 1874 | |
iveresov@2595 | 1875 | if (entry != NULL) { |
iveresov@2595 | 1876 | *entry = __ pc(); |
iveresov@2595 | 1877 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1878 | BLOCK_COMMENT("Entry:"); |
coleenp@548 | 1879 | } |
coleenp@548 | 1880 | |
duke@435 | 1881 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 1882 | // r9 and r10 may be used to save non-volatile registers |
coleenp@548 | 1883 | if (is_oop) { |
coleenp@548 | 1884 | __ movq(saved_to, to); |
iveresov@2606 | 1885 | gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); |
coleenp@548 | 1886 | } |
coleenp@548 | 1887 | |
duke@435 | 1888 | // 'from', 'to' and 'count' are now valid |
never@739 | 1889 | __ movptr(dword_count, count); |
never@739 | 1890 | __ shrptr(count, 1); // count => qword_count |
duke@435 | 1891 | |
duke@435 | 1892 | // Copy from low to high addresses. Use 'to' as scratch. |
never@739 | 1893 | __ lea(end_from, Address(from, qword_count, Address::times_8, -8)); |
never@739 | 1894 | __ lea(end_to, Address(to, qword_count, Address::times_8, -8)); |
never@739 | 1895 | __ negptr(qword_count); |
kvn@4411 | 1896 | __ jmp(L_copy_bytes); |
duke@435 | 1897 | |
duke@435 | 1898 | // Copy trailing qwords |
duke@435 | 1899 | __ BIND(L_copy_8_bytes); |
duke@435 | 1900 | __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); |
duke@435 | 1901 | __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); |
never@739 | 1902 | __ increment(qword_count); |
duke@435 | 1903 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 1904 | |
duke@435 | 1905 | // Check for and copy trailing dword |
duke@435 | 1906 | __ BIND(L_copy_4_bytes); |
never@739 | 1907 | __ testl(dword_count, 1); // Only byte test since the value is 0 or 1 |
duke@435 | 1908 | __ jccb(Assembler::zero, L_exit); |
duke@435 | 1909 | __ movl(rax, Address(end_from, 8)); |
duke@435 | 1910 | __ movl(Address(end_to, 8), rax); |
duke@435 | 1911 | |
duke@435 | 1912 | __ BIND(L_exit); |
coleenp@548 | 1913 | if (is_oop) { |
kvn@5156 | 1914 | gen_write_ref_array_post_barrier(saved_to, dword_count, rax); |
coleenp@548 | 1915 | } |
duke@435 | 1916 | restore_arg_regs(); |
never@3314 | 1917 | inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 1918 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1919 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1920 | __ ret(0); |
duke@435 | 1921 | |
kvn@4411 | 1922 | // Copy in multi-bytes chunks |
kvn@4411 | 1923 | copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); |
duke@435 | 1924 | __ jmp(L_copy_4_bytes); |
duke@435 | 1925 | |
duke@435 | 1926 | return start; |
duke@435 | 1927 | } |
duke@435 | 1928 | |
duke@435 | 1929 | // Arguments: |
duke@435 | 1930 | // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary |
duke@435 | 1931 | // ignored |
coleenp@548 | 1932 | // is_oop - true => oop array, so generate store check code |
duke@435 | 1933 | // name - stub name string |
duke@435 | 1934 | // |
duke@435 | 1935 | // Inputs: |
duke@435 | 1936 | // c_rarg0 - source array address |
duke@435 | 1937 | // c_rarg1 - destination array address |
duke@435 | 1938 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 1939 | // |
duke@435 | 1940 | // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let |
duke@435 | 1941 | // the hardware handle it. The two dwords within qwords that span |
duke@435 | 1942 | // cache line boundaries will still be loaded and stored atomicly. |
duke@435 | 1943 | // |
iveresov@2595 | 1944 | address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target, |
iveresov@2606 | 1945 | address *entry, const char *name, |
iveresov@2606 | 1946 | bool dest_uninitialized = false) { |
duke@435 | 1947 | __ align(CodeEntryAlignment); |
duke@435 | 1948 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1949 | address start = __ pc(); |
duke@435 | 1950 | |
kvn@4411 | 1951 | Label L_copy_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit; |
duke@435 | 1952 | const Register from = rdi; // source array address |
duke@435 | 1953 | const Register to = rsi; // destination array address |
duke@435 | 1954 | const Register count = rdx; // elements count |
duke@435 | 1955 | const Register dword_count = rcx; |
duke@435 | 1956 | const Register qword_count = count; |
duke@435 | 1957 | |
duke@435 | 1958 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1959 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 1960 | |
iveresov@2595 | 1961 | if (entry != NULL) { |
iveresov@2595 | 1962 | *entry = __ pc(); |
iveresov@2595 | 1963 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1964 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1965 | } |
iveresov@2595 | 1966 | |
iveresov@2595 | 1967 | array_overlap_test(nooverlap_target, Address::times_4); |
iveresov@2595 | 1968 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
iveresov@2595 | 1969 | // r9 and r10 may be used to save non-volatile registers |
iveresov@2595 | 1970 | |
coleenp@548 | 1971 | if (is_oop) { |
coleenp@548 | 1972 | // no registers are destroyed by this call |
iveresov@2606 | 1973 | gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); |
coleenp@548 | 1974 | } |
coleenp@548 | 1975 | |
coleenp@548 | 1976 | assert_clean_int(count, rax); // Make sure 'count' is clean int. |
duke@435 | 1977 | // 'from', 'to' and 'count' are now valid |
never@739 | 1978 | __ movptr(dword_count, count); |
never@739 | 1979 | __ shrptr(count, 1); // count => qword_count |
duke@435 | 1980 | |
duke@435 | 1981 | // Copy from high to low addresses. Use 'to' as scratch. |
duke@435 | 1982 | |
duke@435 | 1983 | // Check for and copy trailing dword |
never@739 | 1984 | __ testl(dword_count, 1); |
kvn@4411 | 1985 | __ jcc(Assembler::zero, L_copy_bytes); |
duke@435 | 1986 | __ movl(rax, Address(from, dword_count, Address::times_4, -4)); |
duke@435 | 1987 | __ movl(Address(to, dword_count, Address::times_4, -4), rax); |
kvn@4411 | 1988 | __ jmp(L_copy_bytes); |
duke@435 | 1989 | |
duke@435 | 1990 | // Copy trailing qwords |
duke@435 | 1991 | __ BIND(L_copy_8_bytes); |
duke@435 | 1992 | __ movq(rax, Address(from, qword_count, Address::times_8, -8)); |
duke@435 | 1993 | __ movq(Address(to, qword_count, Address::times_8, -8), rax); |
never@739 | 1994 | __ decrement(qword_count); |
duke@435 | 1995 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 1996 | |
coleenp@548 | 1997 | if (is_oop) { |
coleenp@548 | 1998 | __ jmp(L_exit); |
coleenp@548 | 1999 | } |
duke@435 | 2000 | restore_arg_regs(); |
never@3314 | 2001 | inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 2002 | __ xorptr(rax, rax); // return 0 |
duke@435 | 2003 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2004 | __ ret(0); |
duke@435 | 2005 | |
kvn@4411 | 2006 | // Copy in multi-bytes chunks |
kvn@4411 | 2007 | copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); |
duke@435 | 2008 | |
kvn@5156 | 2009 | __ BIND(L_exit); |
kvn@5156 | 2010 | if (is_oop) { |
kvn@5156 | 2011 | gen_write_ref_array_post_barrier(to, dword_count, rax); |
kvn@5156 | 2012 | } |
duke@435 | 2013 | restore_arg_regs(); |
never@3314 | 2014 | inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 2015 | __ xorptr(rax, rax); // return 0 |
duke@435 | 2016 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2017 | __ ret(0); |
duke@435 | 2018 | |
duke@435 | 2019 | return start; |
duke@435 | 2020 | } |
duke@435 | 2021 | |
duke@435 | 2022 | // Arguments: |
duke@435 | 2023 | // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes |
duke@435 | 2024 | // ignored |
duke@435 | 2025 | // is_oop - true => oop array, so generate store check code |
duke@435 | 2026 | // name - stub name string |
duke@435 | 2027 | // |
duke@435 | 2028 | // Inputs: |
duke@435 | 2029 | // c_rarg0 - source array address |
duke@435 | 2030 | // c_rarg1 - destination array address |
duke@435 | 2031 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 2032 | // |
coleenp@548 | 2033 | // Side Effects: |
duke@435 | 2034 | // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the |
duke@435 | 2035 | // no-overlap entry point used by generate_conjoint_long_oop_copy(). |
duke@435 | 2036 | // |
iveresov@2606 | 2037 | address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry, |
iveresov@2606 | 2038 | const char *name, bool dest_uninitialized = false) { |
duke@435 | 2039 | __ align(CodeEntryAlignment); |
duke@435 | 2040 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2041 | address start = __ pc(); |
duke@435 | 2042 | |
kvn@4411 | 2043 | Label L_copy_bytes, L_copy_8_bytes, L_exit; |
duke@435 | 2044 | const Register from = rdi; // source array address |
duke@435 | 2045 | const Register to = rsi; // destination array address |
duke@435 | 2046 | const Register qword_count = rdx; // elements count |
duke@435 | 2047 | const Register end_from = from; // source array end address |
duke@435 | 2048 | const Register end_to = rcx; // destination array end address |
duke@435 | 2049 | const Register saved_to = to; |
kvn@5156 | 2050 | const Register saved_count = r11; |
duke@435 | 2051 | // End pointers are inclusive, and if count is not zero they point |
duke@435 | 2052 | // to the last unit copied: end_to[0] := end_from[0] |
duke@435 | 2053 | |
duke@435 | 2054 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2055 | // Save no-overlap entry point for generate_conjoint_long_oop_copy() |
duke@435 | 2056 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 2057 | |
iveresov@2595 | 2058 | if (entry != NULL) { |
iveresov@2595 | 2059 | *entry = __ pc(); |
iveresov@2595 | 2060 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 2061 | BLOCK_COMMENT("Entry:"); |
duke@435 | 2062 | } |
duke@435 | 2063 | |
duke@435 | 2064 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 2065 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 2066 | // 'from', 'to' and 'qword_count' are now valid |
iveresov@2595 | 2067 | if (is_oop) { |
kvn@5156 | 2068 | // Save to and count for store barrier |
kvn@5156 | 2069 | __ movptr(saved_count, qword_count); |
iveresov@2595 | 2070 | // no registers are destroyed by this call |
iveresov@2606 | 2071 | gen_write_ref_array_pre_barrier(to, qword_count, dest_uninitialized); |
iveresov@2595 | 2072 | } |
duke@435 | 2073 | |
duke@435 | 2074 | // Copy from low to high addresses. Use 'to' as scratch. |
never@739 | 2075 | __ lea(end_from, Address(from, qword_count, Address::times_8, -8)); |
never@739 | 2076 | __ lea(end_to, Address(to, qword_count, Address::times_8, -8)); |
never@739 | 2077 | __ negptr(qword_count); |
kvn@4411 | 2078 | __ jmp(L_copy_bytes); |
duke@435 | 2079 | |
duke@435 | 2080 | // Copy trailing qwords |
duke@435 | 2081 | __ BIND(L_copy_8_bytes); |
duke@435 | 2082 | __ movq(rax, Address(end_from, qword_count, Address::times_8, 8)); |
duke@435 | 2083 | __ movq(Address(end_to, qword_count, Address::times_8, 8), rax); |
never@739 | 2084 | __ increment(qword_count); |
duke@435 | 2085 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 2086 | |
duke@435 | 2087 | if (is_oop) { |
duke@435 | 2088 | __ jmp(L_exit); |
duke@435 | 2089 | } else { |
duke@435 | 2090 | restore_arg_regs(); |
never@3314 | 2091 | inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 2092 | __ xorptr(rax, rax); // return 0 |
duke@435 | 2093 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2094 | __ ret(0); |
duke@435 | 2095 | } |
duke@435 | 2096 | |
kvn@4411 | 2097 | // Copy in multi-bytes chunks |
kvn@4411 | 2098 | copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); |
duke@435 | 2099 | |
duke@435 | 2100 | if (is_oop) { |
duke@435 | 2101 | __ BIND(L_exit); |
kvn@5156 | 2102 | gen_write_ref_array_post_barrier(saved_to, saved_count, rax); |
duke@435 | 2103 | } |
duke@435 | 2104 | restore_arg_regs(); |
never@3314 | 2105 | if (is_oop) { |
never@3314 | 2106 | inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free |
never@3314 | 2107 | } else { |
never@3314 | 2108 | inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free |
never@3314 | 2109 | } |
never@739 | 2110 | __ xorptr(rax, rax); // return 0 |
duke@435 | 2111 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2112 | __ ret(0); |
duke@435 | 2113 | |
duke@435 | 2114 | return start; |
duke@435 | 2115 | } |
duke@435 | 2116 | |
duke@435 | 2117 | // Arguments: |
duke@435 | 2118 | // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes |
duke@435 | 2119 | // ignored |
duke@435 | 2120 | // is_oop - true => oop array, so generate store check code |
duke@435 | 2121 | // name - stub name string |
duke@435 | 2122 | // |
duke@435 | 2123 | // Inputs: |
duke@435 | 2124 | // c_rarg0 - source array address |
duke@435 | 2125 | // c_rarg1 - destination array address |
duke@435 | 2126 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 2127 | // |
iveresov@2606 | 2128 | address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, |
iveresov@2606 | 2129 | address nooverlap_target, address *entry, |
iveresov@2606 | 2130 | const char *name, bool dest_uninitialized = false) { |
duke@435 | 2131 | __ align(CodeEntryAlignment); |
duke@435 | 2132 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2133 | address start = __ pc(); |
duke@435 | 2134 | |
kvn@4411 | 2135 | Label L_copy_bytes, L_copy_8_bytes, L_exit; |
duke@435 | 2136 | const Register from = rdi; // source array address |
duke@435 | 2137 | const Register to = rsi; // destination array address |
duke@435 | 2138 | const Register qword_count = rdx; // elements count |
duke@435 | 2139 | const Register saved_count = rcx; |
duke@435 | 2140 | |
duke@435 | 2141 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2142 | assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int. |
duke@435 | 2143 | |
iveresov@2595 | 2144 | if (entry != NULL) { |
iveresov@2595 | 2145 | *entry = __ pc(); |
iveresov@2595 | 2146 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 2147 | BLOCK_COMMENT("Entry:"); |
duke@435 | 2148 | } |
iveresov@2595 | 2149 | |
iveresov@2595 | 2150 | array_overlap_test(nooverlap_target, Address::times_8); |
duke@435 | 2151 | setup_arg_regs(); // from => rdi, to => rsi, count => rdx |
duke@435 | 2152 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 2153 | // 'from', 'to' and 'qword_count' are now valid |
duke@435 | 2154 | if (is_oop) { |
duke@435 | 2155 | // Save to and count for store barrier |
never@739 | 2156 | __ movptr(saved_count, qword_count); |
duke@435 | 2157 | // No registers are destroyed by this call |
iveresov@2606 | 2158 | gen_write_ref_array_pre_barrier(to, saved_count, dest_uninitialized); |
duke@435 | 2159 | } |
duke@435 | 2160 | |
kvn@4411 | 2161 | __ jmp(L_copy_bytes); |
duke@435 | 2162 | |
duke@435 | 2163 | // Copy trailing qwords |
duke@435 | 2164 | __ BIND(L_copy_8_bytes); |
duke@435 | 2165 | __ movq(rax, Address(from, qword_count, Address::times_8, -8)); |
duke@435 | 2166 | __ movq(Address(to, qword_count, Address::times_8, -8), rax); |
never@739 | 2167 | __ decrement(qword_count); |
duke@435 | 2168 | __ jcc(Assembler::notZero, L_copy_8_bytes); |
duke@435 | 2169 | |
duke@435 | 2170 | if (is_oop) { |
duke@435 | 2171 | __ jmp(L_exit); |
duke@435 | 2172 | } else { |
duke@435 | 2173 | restore_arg_regs(); |
never@3314 | 2174 | inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free |
never@739 | 2175 | __ xorptr(rax, rax); // return 0 |
duke@435 | 2176 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2177 | __ ret(0); |
duke@435 | 2178 | } |
duke@435 | 2179 | |
kvn@4411 | 2180 | // Copy in multi-bytes chunks |
kvn@4411 | 2181 | copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes); |
duke@435 | 2182 | |
duke@435 | 2183 | if (is_oop) { |
duke@435 | 2184 | __ BIND(L_exit); |
kvn@5156 | 2185 | gen_write_ref_array_post_barrier(to, saved_count, rax); |
duke@435 | 2186 | } |
duke@435 | 2187 | restore_arg_regs(); |
never@3314 | 2188 | if (is_oop) { |
never@3314 | 2189 | inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free |
never@3314 | 2190 | } else { |
never@3314 | 2191 | inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free |
never@3314 | 2192 | } |
never@739 | 2193 | __ xorptr(rax, rax); // return 0 |
duke@435 | 2194 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2195 | __ ret(0); |
duke@435 | 2196 | |
duke@435 | 2197 | return start; |
duke@435 | 2198 | } |
duke@435 | 2199 | |
duke@435 | 2200 | |
duke@435 | 2201 | // Helper for generating a dynamic type check. |
duke@435 | 2202 | // Smashes no registers. |
duke@435 | 2203 | void generate_type_check(Register sub_klass, |
duke@435 | 2204 | Register super_check_offset, |
duke@435 | 2205 | Register super_klass, |
duke@435 | 2206 | Label& L_success) { |
duke@435 | 2207 | assert_different_registers(sub_klass, super_check_offset, super_klass); |
duke@435 | 2208 | |
duke@435 | 2209 | BLOCK_COMMENT("type_check:"); |
duke@435 | 2210 | |
duke@435 | 2211 | Label L_miss; |
duke@435 | 2212 | |
jrose@1079 | 2213 | __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL, |
jrose@1079 | 2214 | super_check_offset); |
jrose@1079 | 2215 | __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL); |
duke@435 | 2216 | |
duke@435 | 2217 | // Fall through on failure! |
duke@435 | 2218 | __ BIND(L_miss); |
duke@435 | 2219 | } |
duke@435 | 2220 | |
duke@435 | 2221 | // |
duke@435 | 2222 | // Generate checkcasting array copy stub |
duke@435 | 2223 | // |
duke@435 | 2224 | // Input: |
duke@435 | 2225 | // c_rarg0 - source array address |
duke@435 | 2226 | // c_rarg1 - destination array address |
duke@435 | 2227 | // c_rarg2 - element count, treated as ssize_t, can be zero |
duke@435 | 2228 | // c_rarg3 - size_t ckoff (super_check_offset) |
duke@435 | 2229 | // not Win64 |
duke@435 | 2230 | // c_rarg4 - oop ckval (super_klass) |
duke@435 | 2231 | // Win64 |
duke@435 | 2232 | // rsp+40 - oop ckval (super_klass) |
duke@435 | 2233 | // |
duke@435 | 2234 | // Output: |
duke@435 | 2235 | // rax == 0 - success |
duke@435 | 2236 | // rax == -1^K - failure, where K is partial transfer count |
duke@435 | 2237 | // |
iveresov@2606 | 2238 | address generate_checkcast_copy(const char *name, address *entry, |
iveresov@2606 | 2239 | bool dest_uninitialized = false) { |
duke@435 | 2240 | |
duke@435 | 2241 | Label L_load_element, L_store_element, L_do_card_marks, L_done; |
duke@435 | 2242 | |
duke@435 | 2243 | // Input registers (after setup_arg_regs) |
duke@435 | 2244 | const Register from = rdi; // source array address |
duke@435 | 2245 | const Register to = rsi; // destination array address |
duke@435 | 2246 | const Register length = rdx; // elements count |
duke@435 | 2247 | const Register ckoff = rcx; // super_check_offset |
duke@435 | 2248 | const Register ckval = r8; // super_klass |
duke@435 | 2249 | |
duke@435 | 2250 | // Registers used as temps (r13, r14 are save-on-entry) |
duke@435 | 2251 | const Register end_from = from; // source array end address |
duke@435 | 2252 | const Register end_to = r13; // destination array end address |
duke@435 | 2253 | const Register count = rdx; // -(count_remaining) |
duke@435 | 2254 | const Register r14_length = r14; // saved copy of length |
duke@435 | 2255 | // End pointers are inclusive, and if length is not zero they point |
duke@435 | 2256 | // to the last unit copied: end_to[0] := end_from[0] |
duke@435 | 2257 | |
duke@435 | 2258 | const Register rax_oop = rax; // actual oop copied |
duke@435 | 2259 | const Register r11_klass = r11; // oop._klass |
duke@435 | 2260 | |
duke@435 | 2261 | //--------------------------------------------------------------- |
duke@435 | 2262 | // Assembler stub will be used for this call to arraycopy |
duke@435 | 2263 | // if the two arrays are subtypes of Object[] but the |
duke@435 | 2264 | // destination array type is not equal to or a supertype |
duke@435 | 2265 | // of the source type. Each element must be separately |
duke@435 | 2266 | // checked. |
duke@435 | 2267 | |
duke@435 | 2268 | __ align(CodeEntryAlignment); |
duke@435 | 2269 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2270 | address start = __ pc(); |
duke@435 | 2271 | |
duke@435 | 2272 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2273 | |
duke@435 | 2274 | #ifdef ASSERT |
duke@435 | 2275 | // caller guarantees that the arrays really are different |
duke@435 | 2276 | // otherwise, we would have to make conjoint checks |
duke@435 | 2277 | { Label L; |
coleenp@548 | 2278 | array_overlap_test(L, TIMES_OOP); |
duke@435 | 2279 | __ stop("checkcast_copy within a single array"); |
duke@435 | 2280 | __ bind(L); |
duke@435 | 2281 | } |
duke@435 | 2282 | #endif //ASSERT |
duke@435 | 2283 | |
duke@435 | 2284 | setup_arg_regs(4); // from => rdi, to => rsi, length => rdx |
duke@435 | 2285 | // ckoff => rcx, ckval => r8 |
duke@435 | 2286 | // r9 and r10 may be used to save non-volatile registers |
duke@435 | 2287 | #ifdef _WIN64 |
duke@435 | 2288 | // last argument (#4) is on stack on Win64 |
twisti@2348 | 2289 | __ movptr(ckval, Address(rsp, 6 * wordSize)); |
duke@435 | 2290 | #endif |
duke@435 | 2291 | |
twisti@2348 | 2292 | // Caller of this entry point must set up the argument registers. |
iveresov@2595 | 2293 | if (entry != NULL) { |
iveresov@2595 | 2294 | *entry = __ pc(); |
iveresov@2595 | 2295 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 2296 | } |
twisti@2348 | 2297 | |
twisti@2348 | 2298 | // allocate spill slots for r13, r14 |
twisti@2348 | 2299 | enum { |
twisti@2348 | 2300 | saved_r13_offset, |
twisti@2348 | 2301 | saved_r14_offset, |
twisti@2348 | 2302 | saved_rbp_offset |
twisti@2348 | 2303 | }; |
twisti@2348 | 2304 | __ subptr(rsp, saved_rbp_offset * wordSize); |
twisti@2348 | 2305 | __ movptr(Address(rsp, saved_r13_offset * wordSize), r13); |
twisti@2348 | 2306 | __ movptr(Address(rsp, saved_r14_offset * wordSize), r14); |
twisti@2348 | 2307 | |
duke@435 | 2308 | // check that int operands are properly extended to size_t |
duke@435 | 2309 | assert_clean_int(length, rax); |
duke@435 | 2310 | assert_clean_int(ckoff, rax); |
duke@435 | 2311 | |
duke@435 | 2312 | #ifdef ASSERT |
duke@435 | 2313 | BLOCK_COMMENT("assert consistent ckoff/ckval"); |
duke@435 | 2314 | // The ckoff and ckval must be mutually consistent, |
duke@435 | 2315 | // even though caller generates both. |
duke@435 | 2316 | { Label L; |
stefank@3391 | 2317 | int sco_offset = in_bytes(Klass::super_check_offset_offset()); |
duke@435 | 2318 | __ cmpl(ckoff, Address(ckval, sco_offset)); |
duke@435 | 2319 | __ jcc(Assembler::equal, L); |
duke@435 | 2320 | __ stop("super_check_offset inconsistent"); |
duke@435 | 2321 | __ bind(L); |
duke@435 | 2322 | } |
duke@435 | 2323 | #endif //ASSERT |
duke@435 | 2324 | |
duke@435 | 2325 | // Loop-invariant addresses. They are exclusive end pointers. |
coleenp@548 | 2326 | Address end_from_addr(from, length, TIMES_OOP, 0); |
coleenp@548 | 2327 | Address end_to_addr(to, length, TIMES_OOP, 0); |
duke@435 | 2328 | // Loop-variant addresses. They assume post-incremented count < 0. |
coleenp@548 | 2329 | Address from_element_addr(end_from, count, TIMES_OOP, 0); |
coleenp@548 | 2330 | Address to_element_addr(end_to, count, TIMES_OOP, 0); |
duke@435 | 2331 | |
iveresov@2606 | 2332 | gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); |
duke@435 | 2333 | |
duke@435 | 2334 | // Copy from low to high addresses, indexed from the end of each array. |
never@739 | 2335 | __ lea(end_from, end_from_addr); |
never@739 | 2336 | __ lea(end_to, end_to_addr); |
never@739 | 2337 | __ movptr(r14_length, length); // save a copy of the length |
never@739 | 2338 | assert(length == count, ""); // else fix next line: |
never@739 | 2339 | __ negptr(count); // negate and test the length |
duke@435 | 2340 | __ jcc(Assembler::notZero, L_load_element); |
duke@435 | 2341 | |
duke@435 | 2342 | // Empty array: Nothing to do. |
never@739 | 2343 | __ xorptr(rax, rax); // return 0 on (trivial) success |
duke@435 | 2344 | __ jmp(L_done); |
duke@435 | 2345 | |
duke@435 | 2346 | // ======== begin loop ======== |
duke@435 | 2347 | // (Loop is rotated; its entry is L_load_element.) |
duke@435 | 2348 | // Loop control: |
duke@435 | 2349 | // for (count = -count; count != 0; count++) |
duke@435 | 2350 | // Base pointers src, dst are biased by 8*(count-1),to last element. |
kvn@1800 | 2351 | __ align(OptoLoopAlignment); |
duke@435 | 2352 | |
duke@435 | 2353 | __ BIND(L_store_element); |
coleenp@548 | 2354 | __ store_heap_oop(to_element_addr, rax_oop); // store the oop |
never@739 | 2355 | __ increment(count); // increment the count toward zero |
duke@435 | 2356 | __ jcc(Assembler::zero, L_do_card_marks); |
duke@435 | 2357 | |
duke@435 | 2358 | // ======== loop entry is here ======== |
duke@435 | 2359 | __ BIND(L_load_element); |
coleenp@548 | 2360 | __ load_heap_oop(rax_oop, from_element_addr); // load the oop |
never@739 | 2361 | __ testptr(rax_oop, rax_oop); |
duke@435 | 2362 | __ jcc(Assembler::zero, L_store_element); |
duke@435 | 2363 | |
coleenp@548 | 2364 | __ load_klass(r11_klass, rax_oop);// query the object klass |
duke@435 | 2365 | generate_type_check(r11_klass, ckoff, ckval, L_store_element); |
duke@435 | 2366 | // ======== end loop ======== |
duke@435 | 2367 | |
duke@435 | 2368 | // It was a real error; we must depend on the caller to finish the job. |
duke@435 | 2369 | // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops. |
duke@435 | 2370 | // Emit GC store barriers for the oops we have copied (r14 + rdx), |
duke@435 | 2371 | // and report their number to the caller. |
kvn@5156 | 2372 | assert_different_registers(rax, r14_length, count, to, end_to, rcx, rscratch1); |
kvn@5156 | 2373 | Label L_post_barrier; |
kvn@5156 | 2374 | __ addptr(r14_length, count); // K = (original - remaining) oops |
kvn@5156 | 2375 | __ movptr(rax, r14_length); // save the value |
kvn@5156 | 2376 | __ notptr(rax); // report (-1^K) to caller (does not affect flags) |
kvn@5156 | 2377 | __ jccb(Assembler::notZero, L_post_barrier); |
kvn@5156 | 2378 | __ jmp(L_done); // K == 0, nothing was copied, skip post barrier |
duke@435 | 2379 | |
duke@435 | 2380 | // Come here on success only. |
duke@435 | 2381 | __ BIND(L_do_card_marks); |
kvn@5156 | 2382 | __ xorptr(rax, rax); // return 0 on success |
kvn@5156 | 2383 | |
kvn@5156 | 2384 | __ BIND(L_post_barrier); |
kvn@5156 | 2385 | gen_write_ref_array_post_barrier(to, r14_length, rscratch1); |
duke@435 | 2386 | |
duke@435 | 2387 | // Common exit point (success or failure). |
duke@435 | 2388 | __ BIND(L_done); |
never@739 | 2389 | __ movptr(r13, Address(rsp, saved_r13_offset * wordSize)); |
never@739 | 2390 | __ movptr(r14, Address(rsp, saved_r14_offset * wordSize)); |
duke@435 | 2391 | restore_arg_regs(); |
never@3314 | 2392 | inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free |
duke@435 | 2393 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2394 | __ ret(0); |
duke@435 | 2395 | |
duke@435 | 2396 | return start; |
duke@435 | 2397 | } |
duke@435 | 2398 | |
duke@435 | 2399 | // |
duke@435 | 2400 | // Generate 'unsafe' array copy stub |
duke@435 | 2401 | // Though just as safe as the other stubs, it takes an unscaled |
duke@435 | 2402 | // size_t argument instead of an element count. |
duke@435 | 2403 | // |
duke@435 | 2404 | // Input: |
duke@435 | 2405 | // c_rarg0 - source array address |
duke@435 | 2406 | // c_rarg1 - destination array address |
duke@435 | 2407 | // c_rarg2 - byte count, treated as ssize_t, can be zero |
duke@435 | 2408 | // |
duke@435 | 2409 | // Examines the alignment of the operands and dispatches |
duke@435 | 2410 | // to a long, int, short, or byte copy loop. |
duke@435 | 2411 | // |
iveresov@2595 | 2412 | address generate_unsafe_copy(const char *name, |
iveresov@2595 | 2413 | address byte_copy_entry, address short_copy_entry, |
iveresov@2595 | 2414 | address int_copy_entry, address long_copy_entry) { |
duke@435 | 2415 | |
duke@435 | 2416 | Label L_long_aligned, L_int_aligned, L_short_aligned; |
duke@435 | 2417 | |
duke@435 | 2418 | // Input registers (before setup_arg_regs) |
duke@435 | 2419 | const Register from = c_rarg0; // source array address |
duke@435 | 2420 | const Register to = c_rarg1; // destination array address |
duke@435 | 2421 | const Register size = c_rarg2; // byte count (size_t) |
duke@435 | 2422 | |
duke@435 | 2423 | // Register used as a temp |
duke@435 | 2424 | const Register bits = rax; // test copy of low bits |
duke@435 | 2425 | |
duke@435 | 2426 | __ align(CodeEntryAlignment); |
duke@435 | 2427 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2428 | address start = __ pc(); |
duke@435 | 2429 | |
duke@435 | 2430 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2431 | |
duke@435 | 2432 | // bump this on entry, not on exit: |
duke@435 | 2433 | inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr); |
duke@435 | 2434 | |
never@739 | 2435 | __ mov(bits, from); |
never@739 | 2436 | __ orptr(bits, to); |
never@739 | 2437 | __ orptr(bits, size); |
duke@435 | 2438 | |
duke@435 | 2439 | __ testb(bits, BytesPerLong-1); |
duke@435 | 2440 | __ jccb(Assembler::zero, L_long_aligned); |
duke@435 | 2441 | |
duke@435 | 2442 | __ testb(bits, BytesPerInt-1); |
duke@435 | 2443 | __ jccb(Assembler::zero, L_int_aligned); |
duke@435 | 2444 | |
duke@435 | 2445 | __ testb(bits, BytesPerShort-1); |
duke@435 | 2446 | __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry)); |
duke@435 | 2447 | |
duke@435 | 2448 | __ BIND(L_short_aligned); |
never@739 | 2449 | __ shrptr(size, LogBytesPerShort); // size => short_count |
duke@435 | 2450 | __ jump(RuntimeAddress(short_copy_entry)); |
duke@435 | 2451 | |
duke@435 | 2452 | __ BIND(L_int_aligned); |
never@739 | 2453 | __ shrptr(size, LogBytesPerInt); // size => int_count |
duke@435 | 2454 | __ jump(RuntimeAddress(int_copy_entry)); |
duke@435 | 2455 | |
duke@435 | 2456 | __ BIND(L_long_aligned); |
never@739 | 2457 | __ shrptr(size, LogBytesPerLong); // size => qword_count |
duke@435 | 2458 | __ jump(RuntimeAddress(long_copy_entry)); |
duke@435 | 2459 | |
duke@435 | 2460 | return start; |
duke@435 | 2461 | } |
duke@435 | 2462 | |
duke@435 | 2463 | // Perform range checks on the proposed arraycopy. |
duke@435 | 2464 | // Kills temp, but nothing else. |
duke@435 | 2465 | // Also, clean the sign bits of src_pos and dst_pos. |
duke@435 | 2466 | void arraycopy_range_checks(Register src, // source array oop (c_rarg0) |
duke@435 | 2467 | Register src_pos, // source position (c_rarg1) |
duke@435 | 2468 | Register dst, // destination array oo (c_rarg2) |
duke@435 | 2469 | Register dst_pos, // destination position (c_rarg3) |
duke@435 | 2470 | Register length, |
duke@435 | 2471 | Register temp, |
duke@435 | 2472 | Label& L_failed) { |
duke@435 | 2473 | BLOCK_COMMENT("arraycopy_range_checks:"); |
duke@435 | 2474 | |
duke@435 | 2475 | // if (src_pos + length > arrayOop(src)->length()) FAIL; |
duke@435 | 2476 | __ movl(temp, length); |
duke@435 | 2477 | __ addl(temp, src_pos); // src_pos + length |
duke@435 | 2478 | __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes())); |
duke@435 | 2479 | __ jcc(Assembler::above, L_failed); |
duke@435 | 2480 | |
duke@435 | 2481 | // if (dst_pos + length > arrayOop(dst)->length()) FAIL; |
duke@435 | 2482 | __ movl(temp, length); |
duke@435 | 2483 | __ addl(temp, dst_pos); // dst_pos + length |
duke@435 | 2484 | __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes())); |
duke@435 | 2485 | __ jcc(Assembler::above, L_failed); |
duke@435 | 2486 | |
duke@435 | 2487 | // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'. |
duke@435 | 2488 | // Move with sign extension can be used since they are positive. |
duke@435 | 2489 | __ movslq(src_pos, src_pos); |
duke@435 | 2490 | __ movslq(dst_pos, dst_pos); |
duke@435 | 2491 | |
duke@435 | 2492 | BLOCK_COMMENT("arraycopy_range_checks done"); |
duke@435 | 2493 | } |
duke@435 | 2494 | |
duke@435 | 2495 | // |
duke@435 | 2496 | // Generate generic array copy stubs |
duke@435 | 2497 | // |
duke@435 | 2498 | // Input: |
duke@435 | 2499 | // c_rarg0 - src oop |
duke@435 | 2500 | // c_rarg1 - src_pos (32-bits) |
duke@435 | 2501 | // c_rarg2 - dst oop |
duke@435 | 2502 | // c_rarg3 - dst_pos (32-bits) |
duke@435 | 2503 | // not Win64 |
duke@435 | 2504 | // c_rarg4 - element count (32-bits) |
duke@435 | 2505 | // Win64 |
duke@435 | 2506 | // rsp+40 - element count (32-bits) |
duke@435 | 2507 | // |
duke@435 | 2508 | // Output: |
duke@435 | 2509 | // rax == 0 - success |
duke@435 | 2510 | // rax == -1^K - failure, where K is partial transfer count |
duke@435 | 2511 | // |
iveresov@2595 | 2512 | address generate_generic_copy(const char *name, |
iveresov@2595 | 2513 | address byte_copy_entry, address short_copy_entry, |
iveresov@2691 | 2514 | address int_copy_entry, address oop_copy_entry, |
iveresov@2691 | 2515 | address long_copy_entry, address checkcast_copy_entry) { |
duke@435 | 2516 | |
duke@435 | 2517 | Label L_failed, L_failed_0, L_objArray; |
duke@435 | 2518 | Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs; |
duke@435 | 2519 | |
duke@435 | 2520 | // Input registers |
duke@435 | 2521 | const Register src = c_rarg0; // source array oop |
duke@435 | 2522 | const Register src_pos = c_rarg1; // source position |
duke@435 | 2523 | const Register dst = c_rarg2; // destination array oop |
duke@435 | 2524 | const Register dst_pos = c_rarg3; // destination position |
twisti@2348 | 2525 | #ifndef _WIN64 |
twisti@2348 | 2526 | const Register length = c_rarg4; |
duke@435 | 2527 | #else |
twisti@2348 | 2528 | const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64 |
duke@435 | 2529 | #endif |
duke@435 | 2530 | |
duke@435 | 2531 | { int modulus = CodeEntryAlignment; |
duke@435 | 2532 | int target = modulus - 5; // 5 = sizeof jmp(L_failed) |
duke@435 | 2533 | int advance = target - (__ offset() % modulus); |
duke@435 | 2534 | if (advance < 0) advance += modulus; |
duke@435 | 2535 | if (advance > 0) __ nop(advance); |
duke@435 | 2536 | } |
duke@435 | 2537 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2538 | |
duke@435 | 2539 | // Short-hop target to L_failed. Makes for denser prologue code. |
duke@435 | 2540 | __ BIND(L_failed_0); |
duke@435 | 2541 | __ jmp(L_failed); |
duke@435 | 2542 | assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed"); |
duke@435 | 2543 | |
duke@435 | 2544 | __ align(CodeEntryAlignment); |
duke@435 | 2545 | address start = __ pc(); |
duke@435 | 2546 | |
duke@435 | 2547 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2548 | |
duke@435 | 2549 | // bump this on entry, not on exit: |
duke@435 | 2550 | inc_counter_np(SharedRuntime::_generic_array_copy_ctr); |
duke@435 | 2551 | |
duke@435 | 2552 | //----------------------------------------------------------------------- |
duke@435 | 2553 | // Assembler stub will be used for this call to arraycopy |
duke@435 | 2554 | // if the following conditions are met: |
duke@435 | 2555 | // |
duke@435 | 2556 | // (1) src and dst must not be null. |
duke@435 | 2557 | // (2) src_pos must not be negative. |
duke@435 | 2558 | // (3) dst_pos must not be negative. |
duke@435 | 2559 | // (4) length must not be negative. |
duke@435 | 2560 | // (5) src klass and dst klass should be the same and not NULL. |
duke@435 | 2561 | // (6) src and dst should be arrays. |
duke@435 | 2562 | // (7) src_pos + length must not exceed length of src. |
duke@435 | 2563 | // (8) dst_pos + length must not exceed length of dst. |
duke@435 | 2564 | // |
duke@435 | 2565 | |
duke@435 | 2566 | // if (src == NULL) return -1; |
never@739 | 2567 | __ testptr(src, src); // src oop |
duke@435 | 2568 | size_t j1off = __ offset(); |
duke@435 | 2569 | __ jccb(Assembler::zero, L_failed_0); |
duke@435 | 2570 | |
duke@435 | 2571 | // if (src_pos < 0) return -1; |
duke@435 | 2572 | __ testl(src_pos, src_pos); // src_pos (32-bits) |
duke@435 | 2573 | __ jccb(Assembler::negative, L_failed_0); |
duke@435 | 2574 | |
duke@435 | 2575 | // if (dst == NULL) return -1; |
never@739 | 2576 | __ testptr(dst, dst); // dst oop |
duke@435 | 2577 | __ jccb(Assembler::zero, L_failed_0); |
duke@435 | 2578 | |
duke@435 | 2579 | // if (dst_pos < 0) return -1; |
duke@435 | 2580 | __ testl(dst_pos, dst_pos); // dst_pos (32-bits) |
duke@435 | 2581 | size_t j4off = __ offset(); |
duke@435 | 2582 | __ jccb(Assembler::negative, L_failed_0); |
duke@435 | 2583 | |
duke@435 | 2584 | // The first four tests are very dense code, |
duke@435 | 2585 | // but not quite dense enough to put four |
duke@435 | 2586 | // jumps in a 16-byte instruction fetch buffer. |
duke@435 | 2587 | // That's good, because some branch predicters |
duke@435 | 2588 | // do not like jumps so close together. |
duke@435 | 2589 | // Make sure of this. |
duke@435 | 2590 | guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps"); |
duke@435 | 2591 | |
duke@435 | 2592 | // registers used as temp |
duke@435 | 2593 | const Register r11_length = r11; // elements count to copy |
duke@435 | 2594 | const Register r10_src_klass = r10; // array klass |
duke@435 | 2595 | |
duke@435 | 2596 | // if (length < 0) return -1; |
twisti@2348 | 2597 | __ movl(r11_length, length); // length (elements count, 32-bits value) |
duke@435 | 2598 | __ testl(r11_length, r11_length); |
duke@435 | 2599 | __ jccb(Assembler::negative, L_failed_0); |
duke@435 | 2600 | |
coleenp@548 | 2601 | __ load_klass(r10_src_klass, src); |
duke@435 | 2602 | #ifdef ASSERT |
duke@435 | 2603 | // assert(src->klass() != NULL); |
twisti@2348 | 2604 | { |
twisti@2348 | 2605 | BLOCK_COMMENT("assert klasses not null {"); |
twisti@2348 | 2606 | Label L1, L2; |
never@739 | 2607 | __ testptr(r10_src_klass, r10_src_klass); |
duke@435 | 2608 | __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL |
duke@435 | 2609 | __ bind(L1); |
duke@435 | 2610 | __ stop("broken null klass"); |
duke@435 | 2611 | __ bind(L2); |
twisti@2348 | 2612 | __ load_klass(rax, dst); |
twisti@2348 | 2613 | __ cmpq(rax, 0); |
duke@435 | 2614 | __ jcc(Assembler::equal, L1); // this would be broken also |
twisti@2348 | 2615 | BLOCK_COMMENT("} assert klasses not null done"); |
duke@435 | 2616 | } |
duke@435 | 2617 | #endif |
duke@435 | 2618 | |
duke@435 | 2619 | // Load layout helper (32-bits) |
duke@435 | 2620 | // |
duke@435 | 2621 | // |array_tag| | header_size | element_type | |log2_element_size| |
duke@435 | 2622 | // 32 30 24 16 8 2 0 |
duke@435 | 2623 | // |
duke@435 | 2624 | // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 |
duke@435 | 2625 | // |
duke@435 | 2626 | |
stefank@3391 | 2627 | const int lh_offset = in_bytes(Klass::layout_helper_offset()); |
twisti@2348 | 2628 | |
twisti@2348 | 2629 | // Handle objArrays completely differently... |
twisti@2348 | 2630 | const jint objArray_lh = Klass::array_layout_helper(T_OBJECT); |
twisti@2348 | 2631 | __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh); |
twisti@2348 | 2632 | __ jcc(Assembler::equal, L_objArray); |
twisti@2348 | 2633 | |
twisti@2348 | 2634 | // if (src->klass() != dst->klass()) return -1; |
twisti@2348 | 2635 | __ load_klass(rax, dst); |
twisti@2348 | 2636 | __ cmpq(r10_src_klass, rax); |
twisti@2348 | 2637 | __ jcc(Assembler::notEqual, L_failed); |
duke@435 | 2638 | |
duke@435 | 2639 | const Register rax_lh = rax; // layout helper |
duke@435 | 2640 | __ movl(rax_lh, Address(r10_src_klass, lh_offset)); |
duke@435 | 2641 | |
duke@435 | 2642 | // if (!src->is_Array()) return -1; |
duke@435 | 2643 | __ cmpl(rax_lh, Klass::_lh_neutral_value); |
duke@435 | 2644 | __ jcc(Assembler::greaterEqual, L_failed); |
duke@435 | 2645 | |
duke@435 | 2646 | // At this point, it is known to be a typeArray (array_tag 0x3). |
duke@435 | 2647 | #ifdef ASSERT |
twisti@2348 | 2648 | { |
twisti@2348 | 2649 | BLOCK_COMMENT("assert primitive array {"); |
twisti@2348 | 2650 | Label L; |
duke@435 | 2651 | __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift)); |
duke@435 | 2652 | __ jcc(Assembler::greaterEqual, L); |
duke@435 | 2653 | __ stop("must be a primitive array"); |
duke@435 | 2654 | __ bind(L); |
twisti@2348 | 2655 | BLOCK_COMMENT("} assert primitive array done"); |
duke@435 | 2656 | } |
duke@435 | 2657 | #endif |
duke@435 | 2658 | |
duke@435 | 2659 | arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, |
duke@435 | 2660 | r10, L_failed); |
duke@435 | 2661 | |
coleenp@4142 | 2662 | // TypeArrayKlass |
duke@435 | 2663 | // |
duke@435 | 2664 | // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); |
duke@435 | 2665 | // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); |
duke@435 | 2666 | // |
duke@435 | 2667 | |
duke@435 | 2668 | const Register r10_offset = r10; // array offset |
duke@435 | 2669 | const Register rax_elsize = rax_lh; // element size |
duke@435 | 2670 | |
duke@435 | 2671 | __ movl(r10_offset, rax_lh); |
duke@435 | 2672 | __ shrl(r10_offset, Klass::_lh_header_size_shift); |
never@739 | 2673 | __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset |
never@739 | 2674 | __ addptr(src, r10_offset); // src array offset |
never@739 | 2675 | __ addptr(dst, r10_offset); // dst array offset |
duke@435 | 2676 | BLOCK_COMMENT("choose copy loop based on element size"); |
duke@435 | 2677 | __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize |
duke@435 | 2678 | |
duke@435 | 2679 | // next registers should be set before the jump to corresponding stub |
duke@435 | 2680 | const Register from = c_rarg0; // source array address |
duke@435 | 2681 | const Register to = c_rarg1; // destination array address |
duke@435 | 2682 | const Register count = c_rarg2; // elements count |
duke@435 | 2683 | |
duke@435 | 2684 | // 'from', 'to', 'count' registers should be set in such order |
duke@435 | 2685 | // since they are the same as 'src', 'src_pos', 'dst'. |
duke@435 | 2686 | |
duke@435 | 2687 | __ BIND(L_copy_bytes); |
duke@435 | 2688 | __ cmpl(rax_elsize, 0); |
duke@435 | 2689 | __ jccb(Assembler::notEqual, L_copy_shorts); |
never@739 | 2690 | __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr |
never@739 | 2691 | __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr |
never@739 | 2692 | __ movl2ptr(count, r11_length); // length |
duke@435 | 2693 | __ jump(RuntimeAddress(byte_copy_entry)); |
duke@435 | 2694 | |
duke@435 | 2695 | __ BIND(L_copy_shorts); |
duke@435 | 2696 | __ cmpl(rax_elsize, LogBytesPerShort); |
duke@435 | 2697 | __ jccb(Assembler::notEqual, L_copy_ints); |
never@739 | 2698 | __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr |
never@739 | 2699 | __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr |
never@739 | 2700 | __ movl2ptr(count, r11_length); // length |
duke@435 | 2701 | __ jump(RuntimeAddress(short_copy_entry)); |
duke@435 | 2702 | |
duke@435 | 2703 | __ BIND(L_copy_ints); |
duke@435 | 2704 | __ cmpl(rax_elsize, LogBytesPerInt); |
duke@435 | 2705 | __ jccb(Assembler::notEqual, L_copy_longs); |
never@739 | 2706 | __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr |
never@739 | 2707 | __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr |
never@739 | 2708 | __ movl2ptr(count, r11_length); // length |
duke@435 | 2709 | __ jump(RuntimeAddress(int_copy_entry)); |
duke@435 | 2710 | |
duke@435 | 2711 | __ BIND(L_copy_longs); |
duke@435 | 2712 | #ifdef ASSERT |
twisti@2348 | 2713 | { |
twisti@2348 | 2714 | BLOCK_COMMENT("assert long copy {"); |
twisti@2348 | 2715 | Label L; |
duke@435 | 2716 | __ cmpl(rax_elsize, LogBytesPerLong); |
duke@435 | 2717 | __ jcc(Assembler::equal, L); |
duke@435 | 2718 | __ stop("must be long copy, but elsize is wrong"); |
duke@435 | 2719 | __ bind(L); |
twisti@2348 | 2720 | BLOCK_COMMENT("} assert long copy done"); |
duke@435 | 2721 | } |
duke@435 | 2722 | #endif |
never@739 | 2723 | __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr |
never@739 | 2724 | __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr |
never@739 | 2725 | __ movl2ptr(count, r11_length); // length |
duke@435 | 2726 | __ jump(RuntimeAddress(long_copy_entry)); |
duke@435 | 2727 | |
coleenp@4142 | 2728 | // ObjArrayKlass |
duke@435 | 2729 | __ BIND(L_objArray); |
twisti@2348 | 2730 | // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos] |
duke@435 | 2731 | |
duke@435 | 2732 | Label L_plain_copy, L_checkcast_copy; |
duke@435 | 2733 | // test array classes for subtyping |
twisti@2348 | 2734 | __ load_klass(rax, dst); |
twisti@2348 | 2735 | __ cmpq(r10_src_klass, rax); // usual case is exact equality |
duke@435 | 2736 | __ jcc(Assembler::notEqual, L_checkcast_copy); |
duke@435 | 2737 | |
duke@435 | 2738 | // Identically typed arrays can be copied without element-wise checks. |
duke@435 | 2739 | arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, |
duke@435 | 2740 | r10, L_failed); |
duke@435 | 2741 | |
never@739 | 2742 | __ lea(from, Address(src, src_pos, TIMES_OOP, |
duke@435 | 2743 | arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr |
never@739 | 2744 | __ lea(to, Address(dst, dst_pos, TIMES_OOP, |
never@739 | 2745 | arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr |
never@739 | 2746 | __ movl2ptr(count, r11_length); // length |
duke@435 | 2747 | __ BIND(L_plain_copy); |
duke@435 | 2748 | __ jump(RuntimeAddress(oop_copy_entry)); |
duke@435 | 2749 | |
duke@435 | 2750 | __ BIND(L_checkcast_copy); |
twisti@2348 | 2751 | // live at this point: r10_src_klass, r11_length, rax (dst_klass) |
duke@435 | 2752 | { |
duke@435 | 2753 | // Before looking at dst.length, make sure dst is also an objArray. |
twisti@2348 | 2754 | __ cmpl(Address(rax, lh_offset), objArray_lh); |
duke@435 | 2755 | __ jcc(Assembler::notEqual, L_failed); |
duke@435 | 2756 | |
duke@435 | 2757 | // It is safe to examine both src.length and dst.length. |
duke@435 | 2758 | arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length, |
duke@435 | 2759 | rax, L_failed); |
twisti@2348 | 2760 | |
twisti@2348 | 2761 | const Register r11_dst_klass = r11; |
coleenp@548 | 2762 | __ load_klass(r11_dst_klass, dst); // reload |
duke@435 | 2763 | |
duke@435 | 2764 | // Marshal the base address arguments now, freeing registers. |
never@739 | 2765 | __ lea(from, Address(src, src_pos, TIMES_OOP, |
duke@435 | 2766 | arrayOopDesc::base_offset_in_bytes(T_OBJECT))); |
never@739 | 2767 | __ lea(to, Address(dst, dst_pos, TIMES_OOP, |
duke@435 | 2768 | arrayOopDesc::base_offset_in_bytes(T_OBJECT))); |
twisti@2348 | 2769 | __ movl(count, length); // length (reloaded) |
duke@435 | 2770 | Register sco_temp = c_rarg3; // this register is free now |
duke@435 | 2771 | assert_different_registers(from, to, count, sco_temp, |
duke@435 | 2772 | r11_dst_klass, r10_src_klass); |
duke@435 | 2773 | assert_clean_int(count, sco_temp); |
duke@435 | 2774 | |
duke@435 | 2775 | // Generate the type check. |
stefank@3391 | 2776 | const int sco_offset = in_bytes(Klass::super_check_offset_offset()); |
duke@435 | 2777 | __ movl(sco_temp, Address(r11_dst_klass, sco_offset)); |
duke@435 | 2778 | assert_clean_int(sco_temp, rax); |
duke@435 | 2779 | generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy); |
duke@435 | 2780 | |
coleenp@4142 | 2781 | // Fetch destination element klass from the ObjArrayKlass header. |
coleenp@4142 | 2782 | int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset()); |
never@739 | 2783 | __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset)); |
twisti@2348 | 2784 | __ movl( sco_temp, Address(r11_dst_klass, sco_offset)); |
duke@435 | 2785 | assert_clean_int(sco_temp, rax); |
duke@435 | 2786 | |
duke@435 | 2787 | // the checkcast_copy loop needs two extra arguments: |
duke@435 | 2788 | assert(c_rarg3 == sco_temp, "#3 already in place"); |
twisti@2348 | 2789 | // Set up arguments for checkcast_copy_entry. |
twisti@2348 | 2790 | setup_arg_regs(4); |
twisti@2348 | 2791 | __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris |
duke@435 | 2792 | __ jump(RuntimeAddress(checkcast_copy_entry)); |
duke@435 | 2793 | } |
duke@435 | 2794 | |
duke@435 | 2795 | __ BIND(L_failed); |
never@739 | 2796 | __ xorptr(rax, rax); |
never@739 | 2797 | __ notptr(rax); // return -1 |
duke@435 | 2798 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 2799 | __ ret(0); |
duke@435 | 2800 | |
duke@435 | 2801 | return start; |
duke@435 | 2802 | } |
duke@435 | 2803 | |
duke@435 | 2804 | void generate_arraycopy_stubs() { |
iveresov@2595 | 2805 | address entry; |
iveresov@2595 | 2806 | address entry_jbyte_arraycopy; |
iveresov@2595 | 2807 | address entry_jshort_arraycopy; |
iveresov@2595 | 2808 | address entry_jint_arraycopy; |
iveresov@2595 | 2809 | address entry_oop_arraycopy; |
iveresov@2595 | 2810 | address entry_jlong_arraycopy; |
iveresov@2595 | 2811 | address entry_checkcast_arraycopy; |
iveresov@2595 | 2812 | |
iveresov@2595 | 2813 | StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry, |
iveresov@2595 | 2814 | "jbyte_disjoint_arraycopy"); |
iveresov@2595 | 2815 | StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy, |
iveresov@2595 | 2816 | "jbyte_arraycopy"); |
iveresov@2595 | 2817 | |
iveresov@2595 | 2818 | StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry, |
iveresov@2595 | 2819 | "jshort_disjoint_arraycopy"); |
iveresov@2595 | 2820 | StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy, |
iveresov@2595 | 2821 | "jshort_arraycopy"); |
iveresov@2595 | 2822 | |
iveresov@2595 | 2823 | StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry, |
iveresov@2595 | 2824 | "jint_disjoint_arraycopy"); |
iveresov@2595 | 2825 | StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry, |
iveresov@2595 | 2826 | &entry_jint_arraycopy, "jint_arraycopy"); |
iveresov@2595 | 2827 | |
iveresov@2595 | 2828 | StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry, |
iveresov@2595 | 2829 | "jlong_disjoint_arraycopy"); |
iveresov@2595 | 2830 | StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry, |
iveresov@2595 | 2831 | &entry_jlong_arraycopy, "jlong_arraycopy"); |
duke@435 | 2832 | |
coleenp@548 | 2833 | |
coleenp@548 | 2834 | if (UseCompressedOops) { |
iveresov@2595 | 2835 | StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry, |
iveresov@2595 | 2836 | "oop_disjoint_arraycopy"); |
iveresov@2595 | 2837 | StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry, |
iveresov@2595 | 2838 | &entry_oop_arraycopy, "oop_arraycopy"); |
iveresov@2606 | 2839 | StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry, |
iveresov@2606 | 2840 | "oop_disjoint_arraycopy_uninit", |
iveresov@2606 | 2841 | /*dest_uninitialized*/true); |
iveresov@2606 | 2842 | StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry, |
iveresov@2606 | 2843 | NULL, "oop_arraycopy_uninit", |
iveresov@2606 | 2844 | /*dest_uninitialized*/true); |
coleenp@548 | 2845 | } else { |
iveresov@2595 | 2846 | StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry, |
iveresov@2595 | 2847 | "oop_disjoint_arraycopy"); |
iveresov@2595 | 2848 | StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry, |
iveresov@2595 | 2849 | &entry_oop_arraycopy, "oop_arraycopy"); |
iveresov@2606 | 2850 | StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry, |
iveresov@2606 | 2851 | "oop_disjoint_arraycopy_uninit", |
iveresov@2606 | 2852 | /*dest_uninitialized*/true); |
iveresov@2606 | 2853 | StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry, |
iveresov@2606 | 2854 | NULL, "oop_arraycopy_uninit", |
iveresov@2606 | 2855 | /*dest_uninitialized*/true); |
coleenp@548 | 2856 | } |
duke@435 | 2857 | |
iveresov@2606 | 2858 | StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy); |
iveresov@2606 | 2859 | StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, |
iveresov@2606 | 2860 | /*dest_uninitialized*/true); |
iveresov@2606 | 2861 | |
iveresov@2595 | 2862 | StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy", |
iveresov@2595 | 2863 | entry_jbyte_arraycopy, |
iveresov@2595 | 2864 | entry_jshort_arraycopy, |
iveresov@2595 | 2865 | entry_jint_arraycopy, |
iveresov@2595 | 2866 | entry_jlong_arraycopy); |
iveresov@2595 | 2867 | StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy", |
iveresov@2595 | 2868 | entry_jbyte_arraycopy, |
iveresov@2595 | 2869 | entry_jshort_arraycopy, |
iveresov@2595 | 2870 | entry_jint_arraycopy, |
iveresov@2595 | 2871 | entry_oop_arraycopy, |
iveresov@2595 | 2872 | entry_jlong_arraycopy, |
iveresov@2595 | 2873 | entry_checkcast_arraycopy); |
duke@435 | 2874 | |
never@2118 | 2875 | StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill"); |
never@2118 | 2876 | StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill"); |
never@2118 | 2877 | StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill"); |
never@2118 | 2878 | StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill"); |
never@2118 | 2879 | StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill"); |
never@2118 | 2880 | StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill"); |
never@2118 | 2881 | |
duke@435 | 2882 | // We don't generate specialized code for HeapWord-aligned source |
duke@435 | 2883 | // arrays, so just use the code we've already generated |
duke@435 | 2884 | StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy; |
duke@435 | 2885 | StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy; |
duke@435 | 2886 | |
duke@435 | 2887 | StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy; |
duke@435 | 2888 | StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy; |
duke@435 | 2889 | |
duke@435 | 2890 | StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy; |
duke@435 | 2891 | StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy; |
duke@435 | 2892 | |
duke@435 | 2893 | StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy; |
duke@435 | 2894 | StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy; |
duke@435 | 2895 | |
duke@435 | 2896 | StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy; |
duke@435 | 2897 | StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; |
iveresov@2606 | 2898 | |
iveresov@2606 | 2899 | StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit; |
iveresov@2606 | 2900 | StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit; |
duke@435 | 2901 | } |
duke@435 | 2902 | |
never@1609 | 2903 | void generate_math_stubs() { |
never@1609 | 2904 | { |
never@1609 | 2905 | StubCodeMark mark(this, "StubRoutines", "log"); |
never@1609 | 2906 | StubRoutines::_intrinsic_log = (double (*)(double)) __ pc(); |
never@1609 | 2907 | |
never@1609 | 2908 | __ subq(rsp, 8); |
never@1609 | 2909 | __ movdbl(Address(rsp, 0), xmm0); |
never@1609 | 2910 | __ fld_d(Address(rsp, 0)); |
never@1609 | 2911 | __ flog(); |
never@1609 | 2912 | __ fstp_d(Address(rsp, 0)); |
never@1609 | 2913 | __ movdbl(xmm0, Address(rsp, 0)); |
never@1609 | 2914 | __ addq(rsp, 8); |
never@1609 | 2915 | __ ret(0); |
never@1609 | 2916 | } |
never@1609 | 2917 | { |
never@1609 | 2918 | StubCodeMark mark(this, "StubRoutines", "log10"); |
never@1609 | 2919 | StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc(); |
never@1609 | 2920 | |
never@1609 | 2921 | __ subq(rsp, 8); |
never@1609 | 2922 | __ movdbl(Address(rsp, 0), xmm0); |
never@1609 | 2923 | __ fld_d(Address(rsp, 0)); |
never@1609 | 2924 | __ flog10(); |
never@1609 | 2925 | __ fstp_d(Address(rsp, 0)); |
never@1609 | 2926 | __ movdbl(xmm0, Address(rsp, 0)); |
never@1609 | 2927 | __ addq(rsp, 8); |
never@1609 | 2928 | __ ret(0); |
never@1609 | 2929 | } |
never@1609 | 2930 | { |
never@1609 | 2931 | StubCodeMark mark(this, "StubRoutines", "sin"); |
never@1609 | 2932 | StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc(); |
never@1609 | 2933 | |
never@1609 | 2934 | __ subq(rsp, 8); |
never@1609 | 2935 | __ movdbl(Address(rsp, 0), xmm0); |
never@1609 | 2936 | __ fld_d(Address(rsp, 0)); |
never@1609 | 2937 | __ trigfunc('s'); |
never@1609 | 2938 | __ fstp_d(Address(rsp, 0)); |
never@1609 | 2939 | __ movdbl(xmm0, Address(rsp, 0)); |
never@1609 | 2940 | __ addq(rsp, 8); |
never@1609 | 2941 | __ ret(0); |
never@1609 | 2942 | } |
never@1609 | 2943 | { |
never@1609 | 2944 | StubCodeMark mark(this, "StubRoutines", "cos"); |
never@1609 | 2945 | StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc(); |
never@1609 | 2946 | |
never@1609 | 2947 | __ subq(rsp, 8); |
never@1609 | 2948 | __ movdbl(Address(rsp, 0), xmm0); |
never@1609 | 2949 | __ fld_d(Address(rsp, 0)); |
never@1609 | 2950 | __ trigfunc('c'); |
never@1609 | 2951 | __ fstp_d(Address(rsp, 0)); |
never@1609 | 2952 | __ movdbl(xmm0, Address(rsp, 0)); |
never@1609 | 2953 | __ addq(rsp, 8); |
never@1609 | 2954 | __ ret(0); |
never@1609 | 2955 | } |
never@1609 | 2956 | { |
never@1609 | 2957 | StubCodeMark mark(this, "StubRoutines", "tan"); |
never@1609 | 2958 | StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc(); |
never@1609 | 2959 | |
never@1609 | 2960 | __ subq(rsp, 8); |
never@1609 | 2961 | __ movdbl(Address(rsp, 0), xmm0); |
never@1609 | 2962 | __ fld_d(Address(rsp, 0)); |
never@1609 | 2963 | __ trigfunc('t'); |
never@1609 | 2964 | __ fstp_d(Address(rsp, 0)); |
never@1609 | 2965 | __ movdbl(xmm0, Address(rsp, 0)); |
never@1609 | 2966 | __ addq(rsp, 8); |
never@1609 | 2967 | __ ret(0); |
never@1609 | 2968 | } |
roland@3787 | 2969 | { |
roland@3787 | 2970 | StubCodeMark mark(this, "StubRoutines", "exp"); |
roland@3787 | 2971 | StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc(); |
roland@3787 | 2972 | |
roland@3787 | 2973 | __ subq(rsp, 8); |
roland@3787 | 2974 | __ movdbl(Address(rsp, 0), xmm0); |
roland@3787 | 2975 | __ fld_d(Address(rsp, 0)); |
roland@3787 | 2976 | __ exp_with_fallback(0); |
roland@3787 | 2977 | __ fstp_d(Address(rsp, 0)); |
roland@3787 | 2978 | __ movdbl(xmm0, Address(rsp, 0)); |
roland@3787 | 2979 | __ addq(rsp, 8); |
roland@3787 | 2980 | __ ret(0); |
roland@3787 | 2981 | } |
roland@3787 | 2982 | { |
roland@3787 | 2983 | StubCodeMark mark(this, "StubRoutines", "pow"); |
roland@3787 | 2984 | StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc(); |
roland@3787 | 2985 | |
roland@3787 | 2986 | __ subq(rsp, 8); |
roland@3787 | 2987 | __ movdbl(Address(rsp, 0), xmm1); |
roland@3787 | 2988 | __ fld_d(Address(rsp, 0)); |
roland@3787 | 2989 | __ movdbl(Address(rsp, 0), xmm0); |
roland@3787 | 2990 | __ fld_d(Address(rsp, 0)); |
roland@3787 | 2991 | __ pow_with_fallback(0); |
roland@3787 | 2992 | __ fstp_d(Address(rsp, 0)); |
roland@3787 | 2993 | __ movdbl(xmm0, Address(rsp, 0)); |
roland@3787 | 2994 | __ addq(rsp, 8); |
roland@3787 | 2995 | __ ret(0); |
roland@3787 | 2996 | } |
never@1609 | 2997 | } |
never@1609 | 2998 | |
kvn@4205 | 2999 | // AES intrinsic stubs |
kvn@4205 | 3000 | enum {AESBlockSize = 16}; |
kvn@4205 | 3001 | |
kvn@4205 | 3002 | address generate_key_shuffle_mask() { |
kvn@4205 | 3003 | __ align(16); |
kvn@4205 | 3004 | StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask"); |
kvn@4205 | 3005 | address start = __ pc(); |
kvn@4205 | 3006 | __ emit_data64( 0x0405060700010203, relocInfo::none ); |
kvn@4205 | 3007 | __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none ); |
kvn@4205 | 3008 | return start; |
kvn@4205 | 3009 | } |
kvn@4205 | 3010 | |
kvn@4205 | 3011 | // Utility routine for loading a 128-bit key word in little endian format |
kvn@4205 | 3012 | // can optionally specify that the shuffle mask is already in an xmmregister |
kvn@4205 | 3013 | void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) { |
kvn@4205 | 3014 | __ movdqu(xmmdst, Address(key, offset)); |
kvn@4205 | 3015 | if (xmm_shuf_mask != NULL) { |
kvn@4205 | 3016 | __ pshufb(xmmdst, xmm_shuf_mask); |
kvn@4205 | 3017 | } else { |
kvn@4205 | 3018 | __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); |
kvn@4205 | 3019 | } |
kvn@4205 | 3020 | } |
kvn@4205 | 3021 | |
kvn@4205 | 3022 | // Arguments: |
kvn@4205 | 3023 | // |
kvn@4205 | 3024 | // Inputs: |
kvn@4205 | 3025 | // c_rarg0 - source byte array address |
kvn@4205 | 3026 | // c_rarg1 - destination byte array address |
kvn@4205 | 3027 | // c_rarg2 - K (key) in little endian int array |
kvn@4205 | 3028 | // |
kvn@4205 | 3029 | address generate_aescrypt_encryptBlock() { |
kvn@4363 | 3030 | assert(UseAES, "need AES instructions and misaligned SSE support"); |
kvn@4205 | 3031 | __ align(CodeEntryAlignment); |
kvn@4205 | 3032 | StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock"); |
kvn@4205 | 3033 | Label L_doLast; |
kvn@4205 | 3034 | address start = __ pc(); |
kvn@4205 | 3035 | |
kvn@4205 | 3036 | const Register from = c_rarg0; // source array address |
kvn@4205 | 3037 | const Register to = c_rarg1; // destination array address |
kvn@4205 | 3038 | const Register key = c_rarg2; // key array address |
kvn@4205 | 3039 | const Register keylen = rax; |
kvn@4205 | 3040 | |
kvn@4205 | 3041 | const XMMRegister xmm_result = xmm0; |
kvn@4363 | 3042 | const XMMRegister xmm_key_shuf_mask = xmm1; |
kvn@4363 | 3043 | // On win64 xmm6-xmm15 must be preserved so don't use them. |
kvn@4363 | 3044 | const XMMRegister xmm_temp1 = xmm2; |
kvn@4363 | 3045 | const XMMRegister xmm_temp2 = xmm3; |
kvn@4363 | 3046 | const XMMRegister xmm_temp3 = xmm4; |
kvn@4363 | 3047 | const XMMRegister xmm_temp4 = xmm5; |
kvn@4205 | 3048 | |
kvn@4205 | 3049 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 3050 | |
kvn@4363 | 3051 | // keylen could be only {11, 13, 15} * 4 = {44, 52, 60} |
kvn@4205 | 3052 | __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); |
kvn@4205 | 3053 | |
kvn@4205 | 3054 | __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); |
kvn@4205 | 3055 | __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input |
kvn@4205 | 3056 | |
kvn@4205 | 3057 | // For encryption, the java expanded key ordering is just what we need |
kvn@4205 | 3058 | // we don't know if the key is aligned, hence not using load-execute form |
kvn@4205 | 3059 | |
kvn@4363 | 3060 | load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask); |
kvn@4363 | 3061 | __ pxor(xmm_result, xmm_temp1); |
kvn@4363 | 3062 | |
kvn@4363 | 3063 | load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask); |
kvn@4363 | 3064 | load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask); |
kvn@4363 | 3065 | load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask); |
kvn@4363 | 3066 | load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask); |
kvn@4363 | 3067 | |
kvn@4363 | 3068 | __ aesenc(xmm_result, xmm_temp1); |
kvn@4363 | 3069 | __ aesenc(xmm_result, xmm_temp2); |
kvn@4363 | 3070 | __ aesenc(xmm_result, xmm_temp3); |
kvn@4363 | 3071 | __ aesenc(xmm_result, xmm_temp4); |
kvn@4363 | 3072 | |
kvn@4363 | 3073 | load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask); |
kvn@4363 | 3074 | load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask); |
kvn@4363 | 3075 | load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask); |
kvn@4363 | 3076 | load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask); |
kvn@4363 | 3077 | |
kvn@4363 | 3078 | __ aesenc(xmm_result, xmm_temp1); |
kvn@4363 | 3079 | __ aesenc(xmm_result, xmm_temp2); |
kvn@4363 | 3080 | __ aesenc(xmm_result, xmm_temp3); |
kvn@4363 | 3081 | __ aesenc(xmm_result, xmm_temp4); |
kvn@4363 | 3082 | |
kvn@4363 | 3083 | load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask); |
kvn@4363 | 3084 | load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask); |
kvn@4363 | 3085 | |
kvn@4363 | 3086 | __ cmpl(keylen, 44); |
kvn@4363 | 3087 | __ jccb(Assembler::equal, L_doLast); |
kvn@4363 | 3088 | |
kvn@4363 | 3089 | __ aesenc(xmm_result, xmm_temp1); |
kvn@4363 | 3090 | __ aesenc(xmm_result, xmm_temp2); |
kvn@4363 | 3091 | |
kvn@4363 | 3092 | load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask); |
kvn@4363 | 3093 | load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask); |
kvn@4363 | 3094 | |
kvn@4363 | 3095 | __ cmpl(keylen, 52); |
kvn@4363 | 3096 | __ jccb(Assembler::equal, L_doLast); |
kvn@4363 | 3097 | |
kvn@4363 | 3098 | __ aesenc(xmm_result, xmm_temp1); |
kvn@4363 | 3099 | __ aesenc(xmm_result, xmm_temp2); |
kvn@4363 | 3100 | |
kvn@4363 | 3101 | load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask); |
kvn@4363 | 3102 | load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask); |
kvn@4205 | 3103 | |
kvn@4205 | 3104 | __ BIND(L_doLast); |
kvn@4363 | 3105 | __ aesenc(xmm_result, xmm_temp1); |
kvn@4363 | 3106 | __ aesenclast(xmm_result, xmm_temp2); |
kvn@4205 | 3107 | __ movdqu(Address(to, 0), xmm_result); // store the result |
kvn@4205 | 3108 | __ xorptr(rax, rax); // return 0 |
kvn@4205 | 3109 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 3110 | __ ret(0); |
kvn@4205 | 3111 | |
kvn@4205 | 3112 | return start; |
kvn@4205 | 3113 | } |
kvn@4205 | 3114 | |
kvn@4205 | 3115 | |
kvn@4205 | 3116 | // Arguments: |
kvn@4205 | 3117 | // |
kvn@4205 | 3118 | // Inputs: |
kvn@4205 | 3119 | // c_rarg0 - source byte array address |
kvn@4205 | 3120 | // c_rarg1 - destination byte array address |
kvn@4205 | 3121 | // c_rarg2 - K (key) in little endian int array |
kvn@4205 | 3122 | // |
kvn@4205 | 3123 | address generate_aescrypt_decryptBlock() { |
kvn@4363 | 3124 | assert(UseAES, "need AES instructions and misaligned SSE support"); |
kvn@4205 | 3125 | __ align(CodeEntryAlignment); |
kvn@4205 | 3126 | StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock"); |
kvn@4205 | 3127 | Label L_doLast; |
kvn@4205 | 3128 | address start = __ pc(); |
kvn@4205 | 3129 | |
kvn@4205 | 3130 | const Register from = c_rarg0; // source array address |
kvn@4205 | 3131 | const Register to = c_rarg1; // destination array address |
kvn@4205 | 3132 | const Register key = c_rarg2; // key array address |
kvn@4205 | 3133 | const Register keylen = rax; |
kvn@4205 | 3134 | |
kvn@4205 | 3135 | const XMMRegister xmm_result = xmm0; |
kvn@4363 | 3136 | const XMMRegister xmm_key_shuf_mask = xmm1; |
kvn@4363 | 3137 | // On win64 xmm6-xmm15 must be preserved so don't use them. |
kvn@4363 | 3138 | const XMMRegister xmm_temp1 = xmm2; |
kvn@4363 | 3139 | const XMMRegister xmm_temp2 = xmm3; |
kvn@4363 | 3140 | const XMMRegister xmm_temp3 = xmm4; |
kvn@4363 | 3141 | const XMMRegister xmm_temp4 = xmm5; |
kvn@4205 | 3142 | |
kvn@4205 | 3143 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 3144 | |
kvn@4363 | 3145 | // keylen could be only {11, 13, 15} * 4 = {44, 52, 60} |
kvn@4205 | 3146 | __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); |
kvn@4205 | 3147 | |
kvn@4205 | 3148 | __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); |
kvn@4205 | 3149 | __ movdqu(xmm_result, Address(from, 0)); |
kvn@4205 | 3150 | |
kvn@4205 | 3151 | // for decryption java expanded key ordering is rotated one position from what we want |
kvn@4205 | 3152 | // so we start from 0x10 here and hit 0x00 last |
kvn@4205 | 3153 | // we don't know if the key is aligned, hence not using load-execute form |
kvn@4363 | 3154 | load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask); |
kvn@4363 | 3155 | load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask); |
kvn@4363 | 3156 | load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask); |
kvn@4363 | 3157 | load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask); |
kvn@4363 | 3158 | |
kvn@4363 | 3159 | __ pxor (xmm_result, xmm_temp1); |
kvn@4363 | 3160 | __ aesdec(xmm_result, xmm_temp2); |
kvn@4363 | 3161 | __ aesdec(xmm_result, xmm_temp3); |
kvn@4363 | 3162 | __ aesdec(xmm_result, xmm_temp4); |
kvn@4363 | 3163 | |
kvn@4363 | 3164 | load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask); |
kvn@4363 | 3165 | load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask); |
kvn@4363 | 3166 | load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask); |
kvn@4363 | 3167 | load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask); |
kvn@4363 | 3168 | |
kvn@4363 | 3169 | __ aesdec(xmm_result, xmm_temp1); |
kvn@4363 | 3170 | __ aesdec(xmm_result, xmm_temp2); |
kvn@4363 | 3171 | __ aesdec(xmm_result, xmm_temp3); |
kvn@4363 | 3172 | __ aesdec(xmm_result, xmm_temp4); |
kvn@4363 | 3173 | |
kvn@4363 | 3174 | load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask); |
kvn@4363 | 3175 | load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask); |
kvn@4363 | 3176 | load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask); |
kvn@4363 | 3177 | |
kvn@4363 | 3178 | __ cmpl(keylen, 44); |
kvn@4363 | 3179 | __ jccb(Assembler::equal, L_doLast); |
kvn@4363 | 3180 | |
kvn@4363 | 3181 | __ aesdec(xmm_result, xmm_temp1); |
kvn@4363 | 3182 | __ aesdec(xmm_result, xmm_temp2); |
kvn@4363 | 3183 | |
kvn@4363 | 3184 | load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask); |
kvn@4363 | 3185 | load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask); |
kvn@4363 | 3186 | |
kvn@4363 | 3187 | __ cmpl(keylen, 52); |
kvn@4363 | 3188 | __ jccb(Assembler::equal, L_doLast); |
kvn@4363 | 3189 | |
kvn@4363 | 3190 | __ aesdec(xmm_result, xmm_temp1); |
kvn@4363 | 3191 | __ aesdec(xmm_result, xmm_temp2); |
kvn@4363 | 3192 | |
kvn@4363 | 3193 | load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask); |
kvn@4363 | 3194 | load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask); |
kvn@4205 | 3195 | |
kvn@4205 | 3196 | __ BIND(L_doLast); |
kvn@4363 | 3197 | __ aesdec(xmm_result, xmm_temp1); |
kvn@4363 | 3198 | __ aesdec(xmm_result, xmm_temp2); |
kvn@4363 | 3199 | |
kvn@4205 | 3200 | // for decryption the aesdeclast operation is always on key+0x00 |
kvn@4363 | 3201 | __ aesdeclast(xmm_result, xmm_temp3); |
kvn@4205 | 3202 | __ movdqu(Address(to, 0), xmm_result); // store the result |
kvn@4205 | 3203 | __ xorptr(rax, rax); // return 0 |
kvn@4205 | 3204 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 3205 | __ ret(0); |
kvn@4205 | 3206 | |
kvn@4205 | 3207 | return start; |
kvn@4205 | 3208 | } |
kvn@4205 | 3209 | |
kvn@4205 | 3210 | |
kvn@4205 | 3211 | // Arguments: |
kvn@4205 | 3212 | // |
kvn@4205 | 3213 | // Inputs: |
kvn@4205 | 3214 | // c_rarg0 - source byte array address |
kvn@4205 | 3215 | // c_rarg1 - destination byte array address |
kvn@4205 | 3216 | // c_rarg2 - K (key) in little endian int array |
kvn@4205 | 3217 | // c_rarg3 - r vector byte array address |
kvn@4205 | 3218 | // c_rarg4 - input length |
kvn@4205 | 3219 | // |
kvn@6312 | 3220 | // Output: |
kvn@6312 | 3221 | // rax - input length |
kvn@6312 | 3222 | // |
kvn@4205 | 3223 | address generate_cipherBlockChaining_encryptAESCrypt() { |
kvn@4363 | 3224 | assert(UseAES, "need AES instructions and misaligned SSE support"); |
kvn@4205 | 3225 | __ align(CodeEntryAlignment); |
kvn@4205 | 3226 | StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt"); |
kvn@4205 | 3227 | address start = __ pc(); |
kvn@4205 | 3228 | |
kvn@4205 | 3229 | Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256; |
kvn@4205 | 3230 | const Register from = c_rarg0; // source array address |
kvn@4205 | 3231 | const Register to = c_rarg1; // destination array address |
kvn@4205 | 3232 | const Register key = c_rarg2; // key array address |
kvn@4205 | 3233 | const Register rvec = c_rarg3; // r byte array initialized from initvector array address |
kvn@4205 | 3234 | // and left with the results of the last encryption block |
kvn@4205 | 3235 | #ifndef _WIN64 |
kvn@4205 | 3236 | const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16) |
kvn@4205 | 3237 | #else |
kvn@6312 | 3238 | const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64 |
kvn@4205 | 3239 | const Register len_reg = r10; // pick the first volatile windows register |
kvn@4205 | 3240 | #endif |
kvn@4205 | 3241 | const Register pos = rax; |
kvn@4205 | 3242 | |
kvn@4205 | 3243 | // xmm register assignments for the loops below |
kvn@4205 | 3244 | const XMMRegister xmm_result = xmm0; |
kvn@4205 | 3245 | const XMMRegister xmm_temp = xmm1; |
kvn@4205 | 3246 | // keys 0-10 preloaded into xmm2-xmm12 |
kvn@4205 | 3247 | const int XMM_REG_NUM_KEY_FIRST = 2; |
kvn@4363 | 3248 | const int XMM_REG_NUM_KEY_LAST = 15; |
kvn@4205 | 3249 | const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST); |
kvn@4363 | 3250 | const XMMRegister xmm_key10 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10); |
kvn@4363 | 3251 | const XMMRegister xmm_key11 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11); |
kvn@4363 | 3252 | const XMMRegister xmm_key12 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12); |
kvn@4363 | 3253 | const XMMRegister xmm_key13 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13); |
kvn@4205 | 3254 | |
kvn@4205 | 3255 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 3256 | |
kvn@4205 | 3257 | #ifdef _WIN64 |
kvn@4205 | 3258 | // on win64, fill len_reg from stack position |
kvn@4205 | 3259 | __ movl(len_reg, len_mem); |
kvn@4363 | 3260 | // save the xmm registers which must be preserved 6-15 |
kvn@4205 | 3261 | __ subptr(rsp, -rsp_after_call_off * wordSize); |
kvn@4205 | 3262 | for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) { |
kvn@4205 | 3263 | __ movdqu(xmm_save(i), as_XMMRegister(i)); |
kvn@4205 | 3264 | } |
kvn@6312 | 3265 | #else |
kvn@6312 | 3266 | __ push(len_reg); // Save |
kvn@4205 | 3267 | #endif |
kvn@4205 | 3268 | |
kvn@4205 | 3269 | const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front |
kvn@4205 | 3270 | __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); |
kvn@4363 | 3271 | // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0 |
kvn@4363 | 3272 | for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) { |
kvn@4205 | 3273 | load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask); |
kvn@4205 | 3274 | offset += 0x10; |
kvn@4205 | 3275 | } |
kvn@4205 | 3276 | __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec |
kvn@4205 | 3277 | |
kvn@4205 | 3278 | // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256)) |
kvn@4205 | 3279 | __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); |
kvn@4205 | 3280 | __ cmpl(rax, 44); |
kvn@4205 | 3281 | __ jcc(Assembler::notEqual, L_key_192_256); |
kvn@4205 | 3282 | |
kvn@4205 | 3283 | // 128 bit code follows here |
kvn@4205 | 3284 | __ movptr(pos, 0); |
kvn@4205 | 3285 | __ align(OptoLoopAlignment); |
kvn@4363 | 3286 | |
kvn@4205 | 3287 | __ BIND(L_loopTop_128); |
kvn@4205 | 3288 | __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input |
kvn@4205 | 3289 | __ pxor (xmm_result, xmm_temp); // xor with the current r vector |
kvn@4205 | 3290 | __ pxor (xmm_result, xmm_key0); // do the aes rounds |
kvn@4363 | 3291 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) { |
kvn@4205 | 3292 | __ aesenc(xmm_result, as_XMMRegister(rnum)); |
kvn@4205 | 3293 | } |
kvn@4205 | 3294 | __ aesenclast(xmm_result, xmm_key10); |
kvn@4205 | 3295 | __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output |
kvn@4205 | 3296 | // no need to store r to memory until we exit |
kvn@4205 | 3297 | __ addptr(pos, AESBlockSize); |
kvn@4205 | 3298 | __ subptr(len_reg, AESBlockSize); |
kvn@4205 | 3299 | __ jcc(Assembler::notEqual, L_loopTop_128); |
kvn@4205 | 3300 | |
kvn@4205 | 3301 | __ BIND(L_exit); |
kvn@4205 | 3302 | __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object |
kvn@4205 | 3303 | |
kvn@4205 | 3304 | #ifdef _WIN64 |
kvn@4205 | 3305 | // restore xmm regs belonging to calling function |
kvn@4205 | 3306 | for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) { |
kvn@4205 | 3307 | __ movdqu(as_XMMRegister(i), xmm_save(i)); |
kvn@4205 | 3308 | } |
kvn@6312 | 3309 | __ movl(rax, len_mem); |
kvn@6312 | 3310 | #else |
kvn@6312 | 3311 | __ pop(rax); // return length |
kvn@4205 | 3312 | #endif |
kvn@4205 | 3313 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 3314 | __ ret(0); |
kvn@4205 | 3315 | |
kvn@4205 | 3316 | __ BIND(L_key_192_256); |
kvn@4205 | 3317 | // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256) |
kvn@4363 | 3318 | load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask); |
kvn@4363 | 3319 | load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask); |
kvn@4205 | 3320 | __ cmpl(rax, 52); |
kvn@4205 | 3321 | __ jcc(Assembler::notEqual, L_key_256); |
kvn@4205 | 3322 | |
kvn@4205 | 3323 | // 192-bit code follows here (could be changed to use more xmm registers) |
kvn@4205 | 3324 | __ movptr(pos, 0); |
kvn@4205 | 3325 | __ align(OptoLoopAlignment); |
kvn@4363 | 3326 | |
kvn@4205 | 3327 | __ BIND(L_loopTop_192); |
kvn@4205 | 3328 | __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input |
kvn@4205 | 3329 | __ pxor (xmm_result, xmm_temp); // xor with the current r vector |
kvn@4205 | 3330 | __ pxor (xmm_result, xmm_key0); // do the aes rounds |
kvn@4363 | 3331 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) { |
kvn@4205 | 3332 | __ aesenc(xmm_result, as_XMMRegister(rnum)); |
kvn@4205 | 3333 | } |
kvn@4363 | 3334 | __ aesenclast(xmm_result, xmm_key12); |
kvn@4205 | 3335 | __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output |
kvn@4205 | 3336 | // no need to store r to memory until we exit |
kvn@4205 | 3337 | __ addptr(pos, AESBlockSize); |
kvn@4205 | 3338 | __ subptr(len_reg, AESBlockSize); |
kvn@4205 | 3339 | __ jcc(Assembler::notEqual, L_loopTop_192); |
kvn@4205 | 3340 | __ jmp(L_exit); |
kvn@4205 | 3341 | |
kvn@4205 | 3342 | __ BIND(L_key_256); |
kvn@4205 | 3343 | // 256-bit code follows here (could be changed to use more xmm registers) |
kvn@4363 | 3344 | load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask); |
kvn@4205 | 3345 | __ movptr(pos, 0); |
kvn@4205 | 3346 | __ align(OptoLoopAlignment); |
kvn@4363 | 3347 | |
kvn@4205 | 3348 | __ BIND(L_loopTop_256); |
kvn@4205 | 3349 | __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input |
kvn@4205 | 3350 | __ pxor (xmm_result, xmm_temp); // xor with the current r vector |
kvn@4205 | 3351 | __ pxor (xmm_result, xmm_key0); // do the aes rounds |
kvn@4363 | 3352 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) { |
kvn@4205 | 3353 | __ aesenc(xmm_result, as_XMMRegister(rnum)); |
kvn@4205 | 3354 | } |
kvn@4205 | 3355 | load_key(xmm_temp, key, 0xe0); |
kvn@4205 | 3356 | __ aesenclast(xmm_result, xmm_temp); |
kvn@4205 | 3357 | __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output |
kvn@4205 | 3358 | // no need to store r to memory until we exit |
kvn@4205 | 3359 | __ addptr(pos, AESBlockSize); |
kvn@4205 | 3360 | __ subptr(len_reg, AESBlockSize); |
kvn@4205 | 3361 | __ jcc(Assembler::notEqual, L_loopTop_256); |
kvn@4205 | 3362 | __ jmp(L_exit); |
kvn@4205 | 3363 | |
kvn@4205 | 3364 | return start; |
kvn@4205 | 3365 | } |
kvn@4205 | 3366 | |
goetz@5400 | 3367 | // Safefetch stubs. |
goetz@5400 | 3368 | void generate_safefetch(const char* name, int size, address* entry, |
goetz@5400 | 3369 | address* fault_pc, address* continuation_pc) { |
goetz@5400 | 3370 | // safefetch signatures: |
goetz@5400 | 3371 | // int SafeFetch32(int* adr, int errValue); |
goetz@5400 | 3372 | // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue); |
goetz@5400 | 3373 | // |
goetz@5400 | 3374 | // arguments: |
goetz@5400 | 3375 | // c_rarg0 = adr |
goetz@5400 | 3376 | // c_rarg1 = errValue |
goetz@5400 | 3377 | // |
goetz@5400 | 3378 | // result: |
goetz@5400 | 3379 | // PPC_RET = *adr or errValue |
goetz@5400 | 3380 | |
goetz@5400 | 3381 | StubCodeMark mark(this, "StubRoutines", name); |
goetz@5400 | 3382 | |
goetz@5400 | 3383 | // Entry point, pc or function descriptor. |
goetz@5400 | 3384 | *entry = __ pc(); |
goetz@5400 | 3385 | |
goetz@5400 | 3386 | // Load *adr into c_rarg1, may fault. |
goetz@5400 | 3387 | *fault_pc = __ pc(); |
goetz@5400 | 3388 | switch (size) { |
goetz@5400 | 3389 | case 4: |
goetz@5400 | 3390 | // int32_t |
goetz@5400 | 3391 | __ movl(c_rarg1, Address(c_rarg0, 0)); |
goetz@5400 | 3392 | break; |
goetz@5400 | 3393 | case 8: |
goetz@5400 | 3394 | // int64_t |
goetz@5400 | 3395 | __ movq(c_rarg1, Address(c_rarg0, 0)); |
goetz@5400 | 3396 | break; |
goetz@5400 | 3397 | default: |
goetz@5400 | 3398 | ShouldNotReachHere(); |
goetz@5400 | 3399 | } |
goetz@5400 | 3400 | |
goetz@5400 | 3401 | // return errValue or *adr |
goetz@5400 | 3402 | *continuation_pc = __ pc(); |
goetz@5400 | 3403 | __ movq(rax, c_rarg1); |
goetz@5400 | 3404 | __ ret(0); |
goetz@5400 | 3405 | } |
kvn@4205 | 3406 | |
kvn@4205 | 3407 | // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time |
kvn@4205 | 3408 | // to hide instruction latency |
kvn@4205 | 3409 | // |
kvn@4205 | 3410 | // Arguments: |
kvn@4205 | 3411 | // |
kvn@4205 | 3412 | // Inputs: |
kvn@4205 | 3413 | // c_rarg0 - source byte array address |
kvn@4205 | 3414 | // c_rarg1 - destination byte array address |
kvn@4205 | 3415 | // c_rarg2 - K (key) in little endian int array |
kvn@4205 | 3416 | // c_rarg3 - r vector byte array address |
kvn@4205 | 3417 | // c_rarg4 - input length |
kvn@4205 | 3418 | // |
kvn@6312 | 3419 | // Output: |
kvn@6312 | 3420 | // rax - input length |
kvn@6312 | 3421 | // |
kvn@4205 | 3422 | |
kvn@4205 | 3423 | address generate_cipherBlockChaining_decryptAESCrypt_Parallel() { |
kvn@4363 | 3424 | assert(UseAES, "need AES instructions and misaligned SSE support"); |
kvn@4205 | 3425 | __ align(CodeEntryAlignment); |
kvn@4205 | 3426 | StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt"); |
kvn@4205 | 3427 | address start = __ pc(); |
kvn@4205 | 3428 | |
kvn@4205 | 3429 | Label L_exit, L_key_192_256, L_key_256; |
kvn@4205 | 3430 | Label L_singleBlock_loopTop_128, L_multiBlock_loopTop_128; |
kvn@4205 | 3431 | Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256; |
kvn@4205 | 3432 | const Register from = c_rarg0; // source array address |
kvn@4205 | 3433 | const Register to = c_rarg1; // destination array address |
kvn@4205 | 3434 | const Register key = c_rarg2; // key array address |
kvn@4205 | 3435 | const Register rvec = c_rarg3; // r byte array initialized from initvector array address |
kvn@4205 | 3436 | // and left with the results of the last encryption block |
kvn@4205 | 3437 | #ifndef _WIN64 |
kvn@4205 | 3438 | const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16) |
kvn@4205 | 3439 | #else |
kvn@6312 | 3440 | const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64 |
kvn@4205 | 3441 | const Register len_reg = r10; // pick the first volatile windows register |
kvn@4205 | 3442 | #endif |
kvn@4205 | 3443 | const Register pos = rax; |
kvn@4205 | 3444 | |
kvn@4205 | 3445 | // keys 0-10 preloaded into xmm2-xmm12 |
kvn@4205 | 3446 | const int XMM_REG_NUM_KEY_FIRST = 5; |
kvn@4205 | 3447 | const int XMM_REG_NUM_KEY_LAST = 15; |
kvn@4363 | 3448 | const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST); |
kvn@4205 | 3449 | const XMMRegister xmm_key_last = as_XMMRegister(XMM_REG_NUM_KEY_LAST); |
kvn@4205 | 3450 | |
kvn@4205 | 3451 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 3452 | |
kvn@4205 | 3453 | #ifdef _WIN64 |
kvn@4205 | 3454 | // on win64, fill len_reg from stack position |
kvn@4205 | 3455 | __ movl(len_reg, len_mem); |
kvn@4205 | 3456 | // save the xmm registers which must be preserved 6-15 |
kvn@4205 | 3457 | __ subptr(rsp, -rsp_after_call_off * wordSize); |
kvn@4205 | 3458 | for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) { |
kvn@4205 | 3459 | __ movdqu(xmm_save(i), as_XMMRegister(i)); |
kvn@4205 | 3460 | } |
kvn@6312 | 3461 | #else |
kvn@6312 | 3462 | __ push(len_reg); // Save |
kvn@4205 | 3463 | #endif |
kvn@6312 | 3464 | |
kvn@4205 | 3465 | // the java expanded key ordering is rotated one position from what we want |
kvn@4205 | 3466 | // so we start from 0x10 here and hit 0x00 last |
kvn@4205 | 3467 | const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front |
kvn@4205 | 3468 | __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); |
kvn@4205 | 3469 | // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00 |
kvn@4363 | 3470 | for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) { |
kvn@4205 | 3471 | load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask); |
kvn@4205 | 3472 | offset += 0x10; |
kvn@4205 | 3473 | } |
kvn@4363 | 3474 | load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask); |
kvn@4205 | 3475 | |
kvn@4205 | 3476 | const XMMRegister xmm_prev_block_cipher = xmm1; // holds cipher of previous block |
kvn@4363 | 3477 | |
kvn@4205 | 3478 | // registers holding the four results in the parallelized loop |
kvn@4205 | 3479 | const XMMRegister xmm_result0 = xmm0; |
kvn@4205 | 3480 | const XMMRegister xmm_result1 = xmm2; |
kvn@4205 | 3481 | const XMMRegister xmm_result2 = xmm3; |
kvn@4205 | 3482 | const XMMRegister xmm_result3 = xmm4; |
kvn@4205 | 3483 | |
kvn@4205 | 3484 | __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec |
kvn@4205 | 3485 | |
kvn@4205 | 3486 | // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256)) |
kvn@4205 | 3487 | __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); |
kvn@4205 | 3488 | __ cmpl(rax, 44); |
kvn@4205 | 3489 | __ jcc(Assembler::notEqual, L_key_192_256); |
kvn@4205 | 3490 | |
kvn@4205 | 3491 | |
kvn@4205 | 3492 | // 128-bit code follows here, parallelized |
kvn@4205 | 3493 | __ movptr(pos, 0); |
kvn@4205 | 3494 | __ align(OptoLoopAlignment); |
kvn@4205 | 3495 | __ BIND(L_multiBlock_loopTop_128); |
kvn@4205 | 3496 | __ cmpptr(len_reg, 4*AESBlockSize); // see if at least 4 blocks left |
kvn@4205 | 3497 | __ jcc(Assembler::less, L_singleBlock_loopTop_128); |
kvn@4205 | 3498 | |
kvn@4205 | 3499 | __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0*AESBlockSize)); // get next 4 blocks into xmmresult registers |
kvn@4205 | 3500 | __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1*AESBlockSize)); |
kvn@4205 | 3501 | __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2*AESBlockSize)); |
kvn@4205 | 3502 | __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3*AESBlockSize)); |
kvn@4205 | 3503 | |
kvn@4205 | 3504 | #define DoFour(opc, src_reg) \ |
kvn@4205 | 3505 | __ opc(xmm_result0, src_reg); \ |
kvn@4205 | 3506 | __ opc(xmm_result1, src_reg); \ |
kvn@4205 | 3507 | __ opc(xmm_result2, src_reg); \ |
kvn@4205 | 3508 | __ opc(xmm_result3, src_reg); |
kvn@4205 | 3509 | |
kvn@4205 | 3510 | DoFour(pxor, xmm_key_first); |
kvn@4205 | 3511 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) { |
kvn@4205 | 3512 | DoFour(aesdec, as_XMMRegister(rnum)); |
kvn@4205 | 3513 | } |
kvn@4205 | 3514 | DoFour(aesdeclast, xmm_key_last); |
kvn@4205 | 3515 | // for each result, xor with the r vector of previous cipher block |
kvn@4205 | 3516 | __ pxor(xmm_result0, xmm_prev_block_cipher); |
kvn@4205 | 3517 | __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0*AESBlockSize)); |
kvn@4205 | 3518 | __ pxor(xmm_result1, xmm_prev_block_cipher); |
kvn@4205 | 3519 | __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1*AESBlockSize)); |
kvn@4205 | 3520 | __ pxor(xmm_result2, xmm_prev_block_cipher); |
kvn@4205 | 3521 | __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2*AESBlockSize)); |
kvn@4205 | 3522 | __ pxor(xmm_result3, xmm_prev_block_cipher); |
kvn@4205 | 3523 | __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3*AESBlockSize)); // this will carry over to next set of blocks |
kvn@4205 | 3524 | |
kvn@4205 | 3525 | __ movdqu(Address(to, pos, Address::times_1, 0*AESBlockSize), xmm_result0); // store 4 results into the next 64 bytes of output |
kvn@4205 | 3526 | __ movdqu(Address(to, pos, Address::times_1, 1*AESBlockSize), xmm_result1); |
kvn@4205 | 3527 | __ movdqu(Address(to, pos, Address::times_1, 2*AESBlockSize), xmm_result2); |
kvn@4205 | 3528 | __ movdqu(Address(to, pos, Address::times_1, 3*AESBlockSize), xmm_result3); |
kvn@4205 | 3529 | |
kvn@4205 | 3530 | __ addptr(pos, 4*AESBlockSize); |
kvn@4205 | 3531 | __ subptr(len_reg, 4*AESBlockSize); |
kvn@4205 | 3532 | __ jmp(L_multiBlock_loopTop_128); |
kvn@4205 | 3533 | |
kvn@4205 | 3534 | // registers used in the non-parallelized loops |
kvn@4363 | 3535 | // xmm register assignments for the loops below |
kvn@4363 | 3536 | const XMMRegister xmm_result = xmm0; |
kvn@4205 | 3537 | const XMMRegister xmm_prev_block_cipher_save = xmm2; |
kvn@4363 | 3538 | const XMMRegister xmm_key11 = xmm3; |
kvn@4363 | 3539 | const XMMRegister xmm_key12 = xmm4; |
kvn@4363 | 3540 | const XMMRegister xmm_temp = xmm4; |
kvn@4205 | 3541 | |
kvn@4205 | 3542 | __ align(OptoLoopAlignment); |
kvn@4205 | 3543 | __ BIND(L_singleBlock_loopTop_128); |
kvn@4205 | 3544 | __ cmpptr(len_reg, 0); // any blocks left?? |
kvn@4205 | 3545 | __ jcc(Assembler::equal, L_exit); |
kvn@4205 | 3546 | __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input |
kvn@4205 | 3547 | __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector |
kvn@4205 | 3548 | __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds |
kvn@4205 | 3549 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) { |
kvn@4205 | 3550 | __ aesdec(xmm_result, as_XMMRegister(rnum)); |
kvn@4205 | 3551 | } |
kvn@4205 | 3552 | __ aesdeclast(xmm_result, xmm_key_last); |
kvn@4205 | 3553 | __ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector |
kvn@4205 | 3554 | __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output |
kvn@4205 | 3555 | // no need to store r to memory until we exit |
kvn@4205 | 3556 | __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block |
kvn@4205 | 3557 | |
kvn@4205 | 3558 | __ addptr(pos, AESBlockSize); |
kvn@4205 | 3559 | __ subptr(len_reg, AESBlockSize); |
kvn@4205 | 3560 | __ jmp(L_singleBlock_loopTop_128); |
kvn@4205 | 3561 | |
kvn@4205 | 3562 | |
kvn@4205 | 3563 | __ BIND(L_exit); |
kvn@4205 | 3564 | __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object |
kvn@4205 | 3565 | #ifdef _WIN64 |
kvn@4205 | 3566 | // restore regs belonging to calling function |
kvn@4205 | 3567 | for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) { |
kvn@4205 | 3568 | __ movdqu(as_XMMRegister(i), xmm_save(i)); |
kvn@4205 | 3569 | } |
kvn@6312 | 3570 | __ movl(rax, len_mem); |
kvn@6312 | 3571 | #else |
kvn@6312 | 3572 | __ pop(rax); // return length |
kvn@4205 | 3573 | #endif |
kvn@4205 | 3574 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 3575 | __ ret(0); |
kvn@4205 | 3576 | |
kvn@4205 | 3577 | |
kvn@4205 | 3578 | __ BIND(L_key_192_256); |
kvn@4205 | 3579 | // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256) |
kvn@4363 | 3580 | load_key(xmm_key11, key, 0xb0); |
kvn@4205 | 3581 | __ cmpl(rax, 52); |
kvn@4205 | 3582 | __ jcc(Assembler::notEqual, L_key_256); |
kvn@4205 | 3583 | |
kvn@4205 | 3584 | // 192-bit code follows here (could be optimized to use parallelism) |
kvn@4363 | 3585 | load_key(xmm_key12, key, 0xc0); // 192-bit key goes up to c0 |
kvn@4205 | 3586 | __ movptr(pos, 0); |
kvn@4205 | 3587 | __ align(OptoLoopAlignment); |
kvn@4363 | 3588 | |
kvn@4205 | 3589 | __ BIND(L_singleBlock_loopTop_192); |
kvn@4205 | 3590 | __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input |
kvn@4205 | 3591 | __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector |
kvn@4205 | 3592 | __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds |
kvn@4205 | 3593 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) { |
kvn@4205 | 3594 | __ aesdec(xmm_result, as_XMMRegister(rnum)); |
kvn@4205 | 3595 | } |
kvn@4363 | 3596 | __ aesdec(xmm_result, xmm_key11); |
kvn@4363 | 3597 | __ aesdec(xmm_result, xmm_key12); |
kvn@4205 | 3598 | __ aesdeclast(xmm_result, xmm_key_last); // xmm15 always came from key+0 |
kvn@4205 | 3599 | __ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector |
kvn@4363 | 3600 | __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output |
kvn@4205 | 3601 | // no need to store r to memory until we exit |
kvn@4363 | 3602 | __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block |
kvn@4205 | 3603 | __ addptr(pos, AESBlockSize); |
kvn@4205 | 3604 | __ subptr(len_reg, AESBlockSize); |
kvn@4205 | 3605 | __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192); |
kvn@4205 | 3606 | __ jmp(L_exit); |
kvn@4205 | 3607 | |
kvn@4205 | 3608 | __ BIND(L_key_256); |
kvn@4205 | 3609 | // 256-bit code follows here (could be optimized to use parallelism) |
kvn@4205 | 3610 | __ movptr(pos, 0); |
kvn@4205 | 3611 | __ align(OptoLoopAlignment); |
kvn@4363 | 3612 | |
kvn@4205 | 3613 | __ BIND(L_singleBlock_loopTop_256); |
kvn@4363 | 3614 | __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input |
kvn@4205 | 3615 | __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector |
kvn@4205 | 3616 | __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds |
kvn@4205 | 3617 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) { |
kvn@4205 | 3618 | __ aesdec(xmm_result, as_XMMRegister(rnum)); |
kvn@4205 | 3619 | } |
kvn@4363 | 3620 | __ aesdec(xmm_result, xmm_key11); |
kvn@4363 | 3621 | load_key(xmm_temp, key, 0xc0); |
kvn@4363 | 3622 | __ aesdec(xmm_result, xmm_temp); |
kvn@4363 | 3623 | load_key(xmm_temp, key, 0xd0); |
kvn@4363 | 3624 | __ aesdec(xmm_result, xmm_temp); |
kvn@4363 | 3625 | load_key(xmm_temp, key, 0xe0); // 256-bit key goes up to e0 |
kvn@4363 | 3626 | __ aesdec(xmm_result, xmm_temp); |
kvn@4363 | 3627 | __ aesdeclast(xmm_result, xmm_key_last); // xmm15 came from key+0 |
kvn@4205 | 3628 | __ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector |
kvn@4363 | 3629 | __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output |
kvn@4205 | 3630 | // no need to store r to memory until we exit |
kvn@4363 | 3631 | __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block |
kvn@4205 | 3632 | __ addptr(pos, AESBlockSize); |
kvn@4205 | 3633 | __ subptr(len_reg, AESBlockSize); |
kvn@4205 | 3634 | __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256); |
kvn@4205 | 3635 | __ jmp(L_exit); |
kvn@4205 | 3636 | |
kvn@4205 | 3637 | return start; |
kvn@4205 | 3638 | } |
kvn@4205 | 3639 | |
drchase@5353 | 3640 | /** |
drchase@5353 | 3641 | * Arguments: |
drchase@5353 | 3642 | * |
drchase@5353 | 3643 | * Inputs: |
drchase@5353 | 3644 | * c_rarg0 - int crc |
drchase@5353 | 3645 | * c_rarg1 - byte* buf |
drchase@5353 | 3646 | * c_rarg2 - int length |
drchase@5353 | 3647 | * |
drchase@5353 | 3648 | * Ouput: |
drchase@5353 | 3649 | * rax - int crc result |
drchase@5353 | 3650 | */ |
drchase@5353 | 3651 | address generate_updateBytesCRC32() { |
drchase@5353 | 3652 | assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions"); |
drchase@5353 | 3653 | |
drchase@5353 | 3654 | __ align(CodeEntryAlignment); |
drchase@5353 | 3655 | StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32"); |
drchase@5353 | 3656 | |
drchase@5353 | 3657 | address start = __ pc(); |
drchase@5353 | 3658 | // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...) |
drchase@5353 | 3659 | // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...) |
drchase@5353 | 3660 | // rscratch1: r10 |
drchase@5353 | 3661 | const Register crc = c_rarg0; // crc |
drchase@5353 | 3662 | const Register buf = c_rarg1; // source java byte array address |
drchase@5353 | 3663 | const Register len = c_rarg2; // length |
drchase@5353 | 3664 | const Register table = c_rarg3; // crc_table address (reuse register) |
drchase@5353 | 3665 | const Register tmp = r11; |
drchase@5353 | 3666 | assert_different_registers(crc, buf, len, table, tmp, rax); |
drchase@5353 | 3667 | |
drchase@5353 | 3668 | BLOCK_COMMENT("Entry:"); |
drchase@5353 | 3669 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
drchase@5353 | 3670 | |
drchase@5353 | 3671 | __ kernel_crc32(crc, buf, len, table, tmp); |
drchase@5353 | 3672 | |
drchase@5353 | 3673 | __ movl(rax, crc); |
drchase@5353 | 3674 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
drchase@5353 | 3675 | __ ret(0); |
drchase@5353 | 3676 | |
drchase@5353 | 3677 | return start; |
drchase@5353 | 3678 | } |
kvn@4205 | 3679 | |
kvn@7152 | 3680 | |
kvn@7152 | 3681 | /** |
kvn@7152 | 3682 | * Arguments: |
kvn@7152 | 3683 | * |
kvn@7152 | 3684 | * Input: |
kvn@7152 | 3685 | * c_rarg0 - x address |
kvn@7152 | 3686 | * c_rarg1 - x length |
kvn@7152 | 3687 | * c_rarg2 - y address |
kvn@7152 | 3688 | * c_rarg3 - y lenth |
kvn@7152 | 3689 | * not Win64 |
kvn@7152 | 3690 | * c_rarg4 - z address |
kvn@7152 | 3691 | * c_rarg5 - z length |
kvn@7152 | 3692 | * Win64 |
kvn@7152 | 3693 | * rsp+40 - z address |
kvn@7152 | 3694 | * rsp+48 - z length |
kvn@7152 | 3695 | */ |
kvn@7152 | 3696 | address generate_multiplyToLen() { |
kvn@7152 | 3697 | __ align(CodeEntryAlignment); |
kvn@7152 | 3698 | StubCodeMark mark(this, "StubRoutines", "multiplyToLen"); |
kvn@7152 | 3699 | |
kvn@7152 | 3700 | address start = __ pc(); |
kvn@7152 | 3701 | // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...) |
kvn@7152 | 3702 | // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...) |
kvn@7152 | 3703 | const Register x = rdi; |
kvn@7152 | 3704 | const Register xlen = rax; |
kvn@7152 | 3705 | const Register y = rsi; |
kvn@7152 | 3706 | const Register ylen = rcx; |
kvn@7152 | 3707 | const Register z = r8; |
kvn@7152 | 3708 | const Register zlen = r11; |
kvn@7152 | 3709 | |
kvn@7152 | 3710 | // Next registers will be saved on stack in multiply_to_len(). |
kvn@7152 | 3711 | const Register tmp1 = r12; |
kvn@7152 | 3712 | const Register tmp2 = r13; |
kvn@7152 | 3713 | const Register tmp3 = r14; |
kvn@7152 | 3714 | const Register tmp4 = r15; |
kvn@7152 | 3715 | const Register tmp5 = rbx; |
kvn@7152 | 3716 | |
kvn@7152 | 3717 | BLOCK_COMMENT("Entry:"); |
kvn@7152 | 3718 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
kvn@7152 | 3719 | |
kvn@7152 | 3720 | #ifndef _WIN64 |
kvn@7152 | 3721 | __ movptr(zlen, r9); // Save r9 in r11 - zlen |
kvn@7152 | 3722 | #endif |
kvn@7152 | 3723 | setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx |
kvn@7152 | 3724 | // ylen => rcx, z => r8, zlen => r11 |
kvn@7152 | 3725 | // r9 and r10 may be used to save non-volatile registers |
kvn@7152 | 3726 | #ifdef _WIN64 |
kvn@7152 | 3727 | // last 2 arguments (#4, #5) are on stack on Win64 |
kvn@7152 | 3728 | __ movptr(z, Address(rsp, 6 * wordSize)); |
kvn@7152 | 3729 | __ movptr(zlen, Address(rsp, 7 * wordSize)); |
kvn@7152 | 3730 | #endif |
kvn@7152 | 3731 | |
kvn@7152 | 3732 | __ movptr(xlen, rsi); |
kvn@7152 | 3733 | __ movptr(y, rdx); |
kvn@7152 | 3734 | __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5); |
kvn@7152 | 3735 | |
kvn@7152 | 3736 | restore_arg_regs(); |
kvn@7152 | 3737 | |
kvn@7152 | 3738 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
kvn@7152 | 3739 | __ ret(0); |
kvn@7152 | 3740 | |
kvn@7152 | 3741 | return start; |
kvn@7152 | 3742 | } |
kvn@7152 | 3743 | |
duke@435 | 3744 | #undef __ |
duke@435 | 3745 | #define __ masm-> |
duke@435 | 3746 | |
duke@435 | 3747 | // Continuation point for throwing of implicit exceptions that are |
duke@435 | 3748 | // not handled in the current activation. Fabricates an exception |
duke@435 | 3749 | // oop and initiates normal exception dispatching in this |
duke@435 | 3750 | // frame. Since we need to preserve callee-saved values (currently |
duke@435 | 3751 | // only for C2, but done for C1 as well) we need a callee-saved oop |
duke@435 | 3752 | // map and therefore have to make these stubs into RuntimeStubs |
duke@435 | 3753 | // rather than BufferBlobs. If the compiler needs all registers to |
duke@435 | 3754 | // be preserved between the fault point and the exception handler |
duke@435 | 3755 | // then it must assume responsibility for that in |
duke@435 | 3756 | // AbstractCompiler::continuation_for_implicit_null_exception or |
duke@435 | 3757 | // continuation_for_implicit_division_by_zero_exception. All other |
duke@435 | 3758 | // implicit exceptions (e.g., NullPointerException or |
duke@435 | 3759 | // AbstractMethodError on entry) are either at call sites or |
duke@435 | 3760 | // otherwise assume that stack unwinding will be initiated, so |
duke@435 | 3761 | // caller saved registers were assumed volatile in the compiler. |
duke@435 | 3762 | address generate_throw_exception(const char* name, |
duke@435 | 3763 | address runtime_entry, |
never@2978 | 3764 | Register arg1 = noreg, |
never@2978 | 3765 | Register arg2 = noreg) { |
duke@435 | 3766 | // Information about frame layout at time of blocking runtime call. |
duke@435 | 3767 | // Note that we only have to preserve callee-saved registers since |
duke@435 | 3768 | // the compilers are responsible for supplying a continuation point |
duke@435 | 3769 | // if they expect all registers to be preserved. |
duke@435 | 3770 | enum layout { |
duke@435 | 3771 | rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt, |
duke@435 | 3772 | rbp_off2, |
duke@435 | 3773 | return_off, |
duke@435 | 3774 | return_off2, |
duke@435 | 3775 | framesize // inclusive of return address |
duke@435 | 3776 | }; |
duke@435 | 3777 | |
duke@435 | 3778 | int insts_size = 512; |
duke@435 | 3779 | int locs_size = 64; |
duke@435 | 3780 | |
duke@435 | 3781 | CodeBuffer code(name, insts_size, locs_size); |
duke@435 | 3782 | OopMapSet* oop_maps = new OopMapSet(); |
duke@435 | 3783 | MacroAssembler* masm = new MacroAssembler(&code); |
duke@435 | 3784 | |
duke@435 | 3785 | address start = __ pc(); |
duke@435 | 3786 | |
duke@435 | 3787 | // This is an inlined and slightly modified version of call_VM |
duke@435 | 3788 | // which has the ability to fetch the return PC out of |
duke@435 | 3789 | // thread-local storage and also sets up last_Java_sp slightly |
duke@435 | 3790 | // differently than the real call_VM |
duke@435 | 3791 | |
duke@435 | 3792 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 3793 | |
duke@435 | 3794 | assert(is_even(framesize/2), "sp not 16-byte aligned"); |
duke@435 | 3795 | |
duke@435 | 3796 | // return address and rbp are already in place |
never@739 | 3797 | __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog |
duke@435 | 3798 | |
duke@435 | 3799 | int frame_complete = __ pc() - start; |
duke@435 | 3800 | |
duke@435 | 3801 | // Set up last_Java_sp and last_Java_fp |
roland@3522 | 3802 | address the_pc = __ pc(); |
roland@3522 | 3803 | __ set_last_Java_frame(rsp, rbp, the_pc); |
roland@3522 | 3804 | __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack |
duke@435 | 3805 | |
duke@435 | 3806 | // Call runtime |
never@2978 | 3807 | if (arg1 != noreg) { |
never@2978 | 3808 | assert(arg2 != c_rarg1, "clobbered"); |
never@2978 | 3809 | __ movptr(c_rarg1, arg1); |
never@2978 | 3810 | } |
never@2978 | 3811 | if (arg2 != noreg) { |
never@2978 | 3812 | __ movptr(c_rarg2, arg2); |
never@2978 | 3813 | } |
never@739 | 3814 | __ movptr(c_rarg0, r15_thread); |
duke@435 | 3815 | BLOCK_COMMENT("call runtime_entry"); |
duke@435 | 3816 | __ call(RuntimeAddress(runtime_entry)); |
duke@435 | 3817 | |
duke@435 | 3818 | // Generate oop map |
duke@435 | 3819 | OopMap* map = new OopMap(framesize, 0); |
duke@435 | 3820 | |
roland@3568 | 3821 | oop_maps->add_gc_map(the_pc - start, map); |
duke@435 | 3822 | |
roland@3522 | 3823 | __ reset_last_Java_frame(true, true); |
duke@435 | 3824 | |
duke@435 | 3825 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 3826 | |
duke@435 | 3827 | // check for pending exceptions |
duke@435 | 3828 | #ifdef ASSERT |
duke@435 | 3829 | Label L; |
never@739 | 3830 | __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), |
never@739 | 3831 | (int32_t) NULL_WORD); |
duke@435 | 3832 | __ jcc(Assembler::notEqual, L); |
duke@435 | 3833 | __ should_not_reach_here(); |
duke@435 | 3834 | __ bind(L); |
duke@435 | 3835 | #endif // ASSERT |
duke@435 | 3836 | __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); |
duke@435 | 3837 | |
duke@435 | 3838 | |
duke@435 | 3839 | // codeBlob framesize is in words (not VMRegImpl::slot_size) |
duke@435 | 3840 | RuntimeStub* stub = |
duke@435 | 3841 | RuntimeStub::new_runtime_stub(name, |
duke@435 | 3842 | &code, |
duke@435 | 3843 | frame_complete, |
duke@435 | 3844 | (framesize >> (LogBytesPerWord - LogBytesPerInt)), |
duke@435 | 3845 | oop_maps, false); |
duke@435 | 3846 | return stub->entry_point(); |
duke@435 | 3847 | } |
duke@435 | 3848 | |
kvn@5439 | 3849 | void create_control_words() { |
kvn@5439 | 3850 | // Round to nearest, 53-bit mode, exceptions masked |
kvn@5439 | 3851 | StubRoutines::_fpu_cntrl_wrd_std = 0x027F; |
kvn@5439 | 3852 | // Round to zero, 53-bit mode, exception mased |
kvn@5439 | 3853 | StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F; |
kvn@5439 | 3854 | // Round to nearest, 24-bit mode, exceptions masked |
kvn@5439 | 3855 | StubRoutines::_fpu_cntrl_wrd_24 = 0x007F; |
kvn@5439 | 3856 | // Round to nearest, 64-bit mode, exceptions masked |
kvn@5439 | 3857 | StubRoutines::_fpu_cntrl_wrd_64 = 0x037F; |
kvn@5439 | 3858 | // Round to nearest, 64-bit mode, exceptions masked |
kvn@5439 | 3859 | StubRoutines::_mxcsr_std = 0x1F80; |
kvn@5439 | 3860 | // Note: the following two constants are 80-bit values |
kvn@5439 | 3861 | // layout is critical for correct loading by FPU. |
kvn@5439 | 3862 | // Bias for strict fp multiply/divide |
kvn@5439 | 3863 | StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000 |
kvn@5439 | 3864 | StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000; |
kvn@5439 | 3865 | StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff; |
kvn@5439 | 3866 | // Un-Bias for strict fp multiply/divide |
kvn@5439 | 3867 | StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000 |
kvn@5439 | 3868 | StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000; |
kvn@5439 | 3869 | StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff; |
kvn@5439 | 3870 | } |
kvn@5439 | 3871 | |
duke@435 | 3872 | // Initialization |
duke@435 | 3873 | void generate_initial() { |
duke@435 | 3874 | // Generates all stubs and initializes the entry points |
duke@435 | 3875 | |
kvn@5439 | 3876 | // This platform-specific settings are needed by generate_call_stub() |
kvn@5439 | 3877 | create_control_words(); |
duke@435 | 3878 | |
duke@435 | 3879 | // entry points that exist in all platforms Note: This is code |
duke@435 | 3880 | // that could be shared among different platforms - however the |
duke@435 | 3881 | // benefit seems to be smaller than the disadvantage of having a |
duke@435 | 3882 | // much more complicated generator structure. See also comment in |
duke@435 | 3883 | // stubRoutines.hpp. |
duke@435 | 3884 | |
duke@435 | 3885 | StubRoutines::_forward_exception_entry = generate_forward_exception(); |
duke@435 | 3886 | |
duke@435 | 3887 | StubRoutines::_call_stub_entry = |
duke@435 | 3888 | generate_call_stub(StubRoutines::_call_stub_return_address); |
duke@435 | 3889 | |
duke@435 | 3890 | // is referenced by megamorphic call |
duke@435 | 3891 | StubRoutines::_catch_exception_entry = generate_catch_exception(); |
duke@435 | 3892 | |
duke@435 | 3893 | // atomic calls |
duke@435 | 3894 | StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); |
duke@435 | 3895 | StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr(); |
duke@435 | 3896 | StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg(); |
duke@435 | 3897 | StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long(); |
duke@435 | 3898 | StubRoutines::_atomic_add_entry = generate_atomic_add(); |
duke@435 | 3899 | StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr(); |
duke@435 | 3900 | StubRoutines::_fence_entry = generate_orderaccess_fence(); |
duke@435 | 3901 | |
duke@435 | 3902 | StubRoutines::_handler_for_unsafe_access_entry = |
duke@435 | 3903 | generate_handler_for_unsafe_access(); |
duke@435 | 3904 | |
duke@435 | 3905 | // platform dependent |
never@739 | 3906 | StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp(); |
roland@3606 | 3907 | StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp(); |
never@739 | 3908 | |
never@739 | 3909 | StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr(); |
never@2978 | 3910 | |
bdelsart@3372 | 3911 | // Build this early so it's available for the interpreter. |
bdelsart@3372 | 3912 | StubRoutines::_throw_StackOverflowError_entry = |
bdelsart@3372 | 3913 | generate_throw_exception("StackOverflowError throw_exception", |
bdelsart@3372 | 3914 | CAST_FROM_FN_PTR(address, |
bdelsart@3372 | 3915 | SharedRuntime:: |
bdelsart@3372 | 3916 | throw_StackOverflowError)); |
drchase@5353 | 3917 | if (UseCRC32Intrinsics) { |
drchase@5353 | 3918 | // set table address before stub generation which use it |
drchase@5353 | 3919 | StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table; |
drchase@5353 | 3920 | StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32(); |
drchase@5353 | 3921 | } |
duke@435 | 3922 | } |
duke@435 | 3923 | |
duke@435 | 3924 | void generate_all() { |
duke@435 | 3925 | // Generates all stubs and initializes the entry points |
duke@435 | 3926 | |
duke@435 | 3927 | // These entry points require SharedInfo::stack0 to be set up in |
duke@435 | 3928 | // non-core builds and need to be relocatable, so they each |
duke@435 | 3929 | // fabricate a RuntimeStub internally. |
duke@435 | 3930 | StubRoutines::_throw_AbstractMethodError_entry = |
duke@435 | 3931 | generate_throw_exception("AbstractMethodError throw_exception", |
duke@435 | 3932 | CAST_FROM_FN_PTR(address, |
duke@435 | 3933 | SharedRuntime:: |
never@3136 | 3934 | throw_AbstractMethodError)); |
duke@435 | 3935 | |
dcubed@451 | 3936 | StubRoutines::_throw_IncompatibleClassChangeError_entry = |
dcubed@451 | 3937 | generate_throw_exception("IncompatibleClassChangeError throw_exception", |
dcubed@451 | 3938 | CAST_FROM_FN_PTR(address, |
dcubed@451 | 3939 | SharedRuntime:: |
never@3136 | 3940 | throw_IncompatibleClassChangeError)); |
duke@435 | 3941 | |
duke@435 | 3942 | StubRoutines::_throw_NullPointerException_at_call_entry = |
duke@435 | 3943 | generate_throw_exception("NullPointerException at call throw_exception", |
duke@435 | 3944 | CAST_FROM_FN_PTR(address, |
duke@435 | 3945 | SharedRuntime:: |
never@3136 | 3946 | throw_NullPointerException_at_call)); |
duke@435 | 3947 | |
duke@435 | 3948 | // entry points that are platform specific |
never@739 | 3949 | StubRoutines::x86::_f2i_fixup = generate_f2i_fixup(); |
never@739 | 3950 | StubRoutines::x86::_f2l_fixup = generate_f2l_fixup(); |
never@739 | 3951 | StubRoutines::x86::_d2i_fixup = generate_d2i_fixup(); |
never@739 | 3952 | StubRoutines::x86::_d2l_fixup = generate_d2l_fixup(); |
never@739 | 3953 | |
never@739 | 3954 | StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF); |
never@739 | 3955 | StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000); |
never@739 | 3956 | StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF); |
never@739 | 3957 | StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000); |
duke@435 | 3958 | |
duke@435 | 3959 | // support for verify_oop (must happen after universe_init) |
duke@435 | 3960 | StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop(); |
duke@435 | 3961 | |
duke@435 | 3962 | // arraycopy stubs used by compilers |
duke@435 | 3963 | generate_arraycopy_stubs(); |
twisti@1543 | 3964 | |
never@1609 | 3965 | generate_math_stubs(); |
kvn@4205 | 3966 | |
kvn@4205 | 3967 | // don't bother generating these AES intrinsic stubs unless global flag is set |
kvn@4205 | 3968 | if (UseAESIntrinsics) { |
kvn@4205 | 3969 | StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // needed by the others |
kvn@4205 | 3970 | |
kvn@4205 | 3971 | StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock(); |
kvn@4205 | 3972 | StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock(); |
kvn@4205 | 3973 | StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt(); |
kvn@4205 | 3974 | StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel(); |
kvn@4205 | 3975 | } |
goetz@5400 | 3976 | |
goetz@5400 | 3977 | // Safefetch stubs. |
goetz@5400 | 3978 | generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry, |
goetz@5400 | 3979 | &StubRoutines::_safefetch32_fault_pc, |
goetz@5400 | 3980 | &StubRoutines::_safefetch32_continuation_pc); |
goetz@5400 | 3981 | generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry, |
goetz@5400 | 3982 | &StubRoutines::_safefetchN_fault_pc, |
goetz@5400 | 3983 | &StubRoutines::_safefetchN_continuation_pc); |
kvn@7152 | 3984 | #ifdef COMPILER2 |
kvn@7152 | 3985 | if (UseMultiplyToLenIntrinsic) { |
kvn@7152 | 3986 | StubRoutines::_multiplyToLen = generate_multiplyToLen(); |
kvn@7152 | 3987 | } |
kvn@7152 | 3988 | #endif |
duke@435 | 3989 | } |
duke@435 | 3990 | |
duke@435 | 3991 | public: |
duke@435 | 3992 | StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { |
duke@435 | 3993 | if (all) { |
duke@435 | 3994 | generate_all(); |
duke@435 | 3995 | } else { |
duke@435 | 3996 | generate_initial(); |
duke@435 | 3997 | } |
duke@435 | 3998 | } |
duke@435 | 3999 | }; // end class declaration |
duke@435 | 4000 | |
duke@435 | 4001 | void StubGenerator_generate(CodeBuffer* code, bool all) { |
duke@435 | 4002 | StubGenerator g(code, all); |
duke@435 | 4003 | } |