Sat, 07 Nov 2020 10:30:02 +0800
Added tag mips-jdk8u275-b01 for changeset d3b4d62f391f
duke@435 | 1 | /* |
drchase@5353 | 2 | * Copyright (c) 1999, 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-> |
never@739 | 51 | #define a__ ((Assembler*)_masm)-> |
duke@435 | 52 | |
duke@435 | 53 | #ifdef PRODUCT |
duke@435 | 54 | #define BLOCK_COMMENT(str) /* nothing */ |
duke@435 | 55 | #else |
duke@435 | 56 | #define BLOCK_COMMENT(str) __ block_comment(str) |
duke@435 | 57 | #endif |
duke@435 | 58 | |
duke@435 | 59 | #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") |
duke@435 | 60 | |
duke@435 | 61 | const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions |
duke@435 | 62 | const int FPU_CNTRL_WRD_MASK = 0xFFFF; |
duke@435 | 63 | |
duke@435 | 64 | // ------------------------------------------------------------------------------------------------------------------------- |
duke@435 | 65 | // Stub Code definitions |
duke@435 | 66 | |
duke@435 | 67 | static address handle_unsafe_access() { |
duke@435 | 68 | JavaThread* thread = JavaThread::current(); |
duke@435 | 69 | address pc = thread->saved_exception_pc(); |
duke@435 | 70 | // pc is the instruction which we must emulate |
duke@435 | 71 | // doing a no-op is fine: return garbage from the load |
duke@435 | 72 | // therefore, compute npc |
duke@435 | 73 | address npc = Assembler::locate_next_instruction(pc); |
duke@435 | 74 | |
duke@435 | 75 | // request an async exception |
duke@435 | 76 | thread->set_pending_unsafe_access_error(); |
duke@435 | 77 | |
duke@435 | 78 | // return address of next instruction to execute |
duke@435 | 79 | return npc; |
duke@435 | 80 | } |
duke@435 | 81 | |
duke@435 | 82 | class StubGenerator: public StubCodeGenerator { |
duke@435 | 83 | private: |
duke@435 | 84 | |
duke@435 | 85 | #ifdef PRODUCT |
ccheung@5259 | 86 | #define inc_counter_np(counter) ((void)0) |
duke@435 | 87 | #else |
duke@435 | 88 | void inc_counter_np_(int& counter) { |
never@739 | 89 | __ incrementl(ExternalAddress((address)&counter)); |
duke@435 | 90 | } |
duke@435 | 91 | #define inc_counter_np(counter) \ |
duke@435 | 92 | BLOCK_COMMENT("inc_counter " #counter); \ |
duke@435 | 93 | inc_counter_np_(counter); |
duke@435 | 94 | #endif //PRODUCT |
duke@435 | 95 | |
duke@435 | 96 | void inc_copy_counter_np(BasicType t) { |
duke@435 | 97 | #ifndef PRODUCT |
duke@435 | 98 | switch (t) { |
duke@435 | 99 | case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return; |
duke@435 | 100 | case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return; |
duke@435 | 101 | case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return; |
duke@435 | 102 | case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return; |
duke@435 | 103 | case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return; |
duke@435 | 104 | } |
duke@435 | 105 | ShouldNotReachHere(); |
duke@435 | 106 | #endif //PRODUCT |
duke@435 | 107 | } |
duke@435 | 108 | |
duke@435 | 109 | //------------------------------------------------------------------------------------------------------------------------ |
duke@435 | 110 | // Call stubs are used to call Java from C |
duke@435 | 111 | // |
duke@435 | 112 | // [ return_from_Java ] <--- rsp |
duke@435 | 113 | // [ argument word n ] |
duke@435 | 114 | // ... |
duke@435 | 115 | // -N [ argument word 1 ] |
duke@435 | 116 | // -7 [ Possible padding for stack alignment ] |
duke@435 | 117 | // -6 [ Possible padding for stack alignment ] |
duke@435 | 118 | // -5 [ Possible padding for stack alignment ] |
duke@435 | 119 | // -4 [ mxcsr save ] <--- rsp_after_call |
duke@435 | 120 | // -3 [ saved rbx, ] |
duke@435 | 121 | // -2 [ saved rsi ] |
duke@435 | 122 | // -1 [ saved rdi ] |
duke@435 | 123 | // 0 [ saved rbp, ] <--- rbp, |
duke@435 | 124 | // 1 [ return address ] |
duke@435 | 125 | // 2 [ ptr. to call wrapper ] |
duke@435 | 126 | // 3 [ result ] |
duke@435 | 127 | // 4 [ result_type ] |
duke@435 | 128 | // 5 [ method ] |
duke@435 | 129 | // 6 [ entry_point ] |
duke@435 | 130 | // 7 [ parameters ] |
duke@435 | 131 | // 8 [ parameter_size ] |
duke@435 | 132 | // 9 [ thread ] |
duke@435 | 133 | |
duke@435 | 134 | |
duke@435 | 135 | address generate_call_stub(address& return_address) { |
duke@435 | 136 | StubCodeMark mark(this, "StubRoutines", "call_stub"); |
duke@435 | 137 | address start = __ pc(); |
duke@435 | 138 | |
duke@435 | 139 | // stub code parameters / addresses |
duke@435 | 140 | assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code"); |
duke@435 | 141 | bool sse_save = false; |
duke@435 | 142 | const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()! |
duke@435 | 143 | const int locals_count_in_bytes (4*wordSize); |
duke@435 | 144 | const Address mxcsr_save (rbp, -4 * wordSize); |
duke@435 | 145 | const Address saved_rbx (rbp, -3 * wordSize); |
duke@435 | 146 | const Address saved_rsi (rbp, -2 * wordSize); |
duke@435 | 147 | const Address saved_rdi (rbp, -1 * wordSize); |
duke@435 | 148 | const Address result (rbp, 3 * wordSize); |
duke@435 | 149 | const Address result_type (rbp, 4 * wordSize); |
duke@435 | 150 | const Address method (rbp, 5 * wordSize); |
duke@435 | 151 | const Address entry_point (rbp, 6 * wordSize); |
duke@435 | 152 | const Address parameters (rbp, 7 * wordSize); |
duke@435 | 153 | const Address parameter_size(rbp, 8 * wordSize); |
duke@435 | 154 | const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()! |
duke@435 | 155 | sse_save = UseSSE > 0; |
duke@435 | 156 | |
duke@435 | 157 | // stub code |
duke@435 | 158 | __ enter(); |
never@739 | 159 | __ movptr(rcx, parameter_size); // parameter counter |
twisti@1861 | 160 | __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes |
never@739 | 161 | __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves |
never@739 | 162 | __ subptr(rsp, rcx); |
never@739 | 163 | __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack |
duke@435 | 164 | |
duke@435 | 165 | // save rdi, rsi, & rbx, according to C calling conventions |
never@739 | 166 | __ movptr(saved_rdi, rdi); |
never@739 | 167 | __ movptr(saved_rsi, rsi); |
never@739 | 168 | __ movptr(saved_rbx, rbx); |
duke@435 | 169 | // save and initialize %mxcsr |
duke@435 | 170 | if (sse_save) { |
duke@435 | 171 | Label skip_ldmx; |
duke@435 | 172 | __ stmxcsr(mxcsr_save); |
duke@435 | 173 | __ movl(rax, mxcsr_save); |
duke@435 | 174 | __ andl(rax, MXCSR_MASK); // Only check control and mask bits |
duke@435 | 175 | ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); |
duke@435 | 176 | __ cmp32(rax, mxcsr_std); |
duke@435 | 177 | __ jcc(Assembler::equal, skip_ldmx); |
duke@435 | 178 | __ ldmxcsr(mxcsr_std); |
duke@435 | 179 | __ bind(skip_ldmx); |
duke@435 | 180 | } |
duke@435 | 181 | |
duke@435 | 182 | // make sure the control word is correct. |
duke@435 | 183 | __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); |
duke@435 | 184 | |
duke@435 | 185 | #ifdef ASSERT |
duke@435 | 186 | // make sure we have no pending exceptions |
duke@435 | 187 | { Label L; |
never@739 | 188 | __ movptr(rcx, thread); |
never@739 | 189 | __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
duke@435 | 190 | __ jcc(Assembler::equal, L); |
duke@435 | 191 | __ stop("StubRoutines::call_stub: entered with pending exception"); |
duke@435 | 192 | __ bind(L); |
duke@435 | 193 | } |
duke@435 | 194 | #endif |
duke@435 | 195 | |
duke@435 | 196 | // pass parameters if any |
duke@435 | 197 | BLOCK_COMMENT("pass parameters if any"); |
duke@435 | 198 | Label parameters_done; |
duke@435 | 199 | __ movl(rcx, parameter_size); // parameter counter |
duke@435 | 200 | __ testl(rcx, rcx); |
duke@435 | 201 | __ jcc(Assembler::zero, parameters_done); |
duke@435 | 202 | |
duke@435 | 203 | // parameter passing loop |
duke@435 | 204 | |
duke@435 | 205 | Label loop; |
duke@435 | 206 | // Copy Java parameters in reverse order (receiver last) |
duke@435 | 207 | // Note that the argument order is inverted in the process |
duke@435 | 208 | // source is rdx[rcx: N-1..0] |
duke@435 | 209 | // dest is rsp[rbx: 0..N-1] |
duke@435 | 210 | |
never@739 | 211 | __ movptr(rdx, parameters); // parameter pointer |
never@739 | 212 | __ xorptr(rbx, rbx); |
duke@435 | 213 | |
duke@435 | 214 | __ BIND(loop); |
duke@435 | 215 | |
duke@435 | 216 | // get parameter |
never@739 | 217 | __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize)); |
never@739 | 218 | __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(), |
duke@435 | 219 | Interpreter::expr_offset_in_bytes(0)), rax); // store parameter |
duke@435 | 220 | __ increment(rbx); |
duke@435 | 221 | __ decrement(rcx); |
duke@435 | 222 | __ jcc(Assembler::notZero, loop); |
duke@435 | 223 | |
duke@435 | 224 | // call Java function |
duke@435 | 225 | __ BIND(parameters_done); |
coleenp@4037 | 226 | __ movptr(rbx, method); // get Method* |
never@739 | 227 | __ movptr(rax, entry_point); // get entry_point |
never@739 | 228 | __ mov(rsi, rsp); // set sender sp |
duke@435 | 229 | BLOCK_COMMENT("call Java function"); |
duke@435 | 230 | __ call(rax); |
duke@435 | 231 | |
duke@435 | 232 | BLOCK_COMMENT("call_stub_return_address:"); |
duke@435 | 233 | return_address = __ pc(); |
duke@435 | 234 | |
twisti@2552 | 235 | #ifdef COMPILER2 |
twisti@2552 | 236 | { |
twisti@2552 | 237 | Label L_skip; |
twisti@2552 | 238 | if (UseSSE >= 2) { |
twisti@2552 | 239 | __ verify_FPU(0, "call_stub_return"); |
twisti@2552 | 240 | } else { |
twisti@2552 | 241 | for (int i = 1; i < 8; i++) { |
twisti@2552 | 242 | __ ffree(i); |
twisti@2552 | 243 | } |
duke@435 | 244 | |
twisti@2552 | 245 | // UseSSE <= 1 so double result should be left on TOS |
twisti@2552 | 246 | __ movl(rsi, result_type); |
twisti@2552 | 247 | __ cmpl(rsi, T_DOUBLE); |
twisti@2552 | 248 | __ jcc(Assembler::equal, L_skip); |
twisti@2552 | 249 | if (UseSSE == 0) { |
twisti@2552 | 250 | // UseSSE == 0 so float result should be left on TOS |
twisti@2552 | 251 | __ cmpl(rsi, T_FLOAT); |
twisti@2552 | 252 | __ jcc(Assembler::equal, L_skip); |
twisti@2552 | 253 | } |
twisti@2552 | 254 | __ ffree(0); |
twisti@2552 | 255 | } |
twisti@2552 | 256 | __ BIND(L_skip); |
twisti@2552 | 257 | } |
twisti@2552 | 258 | #endif // COMPILER2 |
duke@435 | 259 | |
duke@435 | 260 | // store result depending on type |
duke@435 | 261 | // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT) |
never@739 | 262 | __ movptr(rdi, result); |
duke@435 | 263 | Label is_long, is_float, is_double, exit; |
duke@435 | 264 | __ movl(rsi, result_type); |
duke@435 | 265 | __ cmpl(rsi, T_LONG); |
duke@435 | 266 | __ jcc(Assembler::equal, is_long); |
duke@435 | 267 | __ cmpl(rsi, T_FLOAT); |
duke@435 | 268 | __ jcc(Assembler::equal, is_float); |
duke@435 | 269 | __ cmpl(rsi, T_DOUBLE); |
duke@435 | 270 | __ jcc(Assembler::equal, is_double); |
duke@435 | 271 | |
duke@435 | 272 | // handle T_INT case |
duke@435 | 273 | __ movl(Address(rdi, 0), rax); |
duke@435 | 274 | __ BIND(exit); |
duke@435 | 275 | |
duke@435 | 276 | // check that FPU stack is empty |
duke@435 | 277 | __ verify_FPU(0, "generate_call_stub"); |
duke@435 | 278 | |
duke@435 | 279 | // pop parameters |
never@739 | 280 | __ lea(rsp, rsp_after_call); |
duke@435 | 281 | |
duke@435 | 282 | // restore %mxcsr |
duke@435 | 283 | if (sse_save) { |
duke@435 | 284 | __ ldmxcsr(mxcsr_save); |
duke@435 | 285 | } |
duke@435 | 286 | |
duke@435 | 287 | // restore rdi, rsi and rbx, |
never@739 | 288 | __ movptr(rbx, saved_rbx); |
never@739 | 289 | __ movptr(rsi, saved_rsi); |
never@739 | 290 | __ movptr(rdi, saved_rdi); |
never@739 | 291 | __ addptr(rsp, 4*wordSize); |
duke@435 | 292 | |
duke@435 | 293 | // return |
never@739 | 294 | __ pop(rbp); |
duke@435 | 295 | __ ret(0); |
duke@435 | 296 | |
duke@435 | 297 | // handle return types different from T_INT |
duke@435 | 298 | __ BIND(is_long); |
duke@435 | 299 | __ movl(Address(rdi, 0 * wordSize), rax); |
duke@435 | 300 | __ movl(Address(rdi, 1 * wordSize), rdx); |
duke@435 | 301 | __ jmp(exit); |
duke@435 | 302 | |
duke@435 | 303 | __ BIND(is_float); |
duke@435 | 304 | // interpreter uses xmm0 for return values |
duke@435 | 305 | if (UseSSE >= 1) { |
duke@435 | 306 | __ movflt(Address(rdi, 0), xmm0); |
duke@435 | 307 | } else { |
duke@435 | 308 | __ fstp_s(Address(rdi, 0)); |
duke@435 | 309 | } |
duke@435 | 310 | __ jmp(exit); |
duke@435 | 311 | |
duke@435 | 312 | __ BIND(is_double); |
duke@435 | 313 | // interpreter uses xmm0 for return values |
duke@435 | 314 | if (UseSSE >= 2) { |
duke@435 | 315 | __ movdbl(Address(rdi, 0), xmm0); |
duke@435 | 316 | } else { |
duke@435 | 317 | __ fstp_d(Address(rdi, 0)); |
duke@435 | 318 | } |
duke@435 | 319 | __ jmp(exit); |
duke@435 | 320 | |
duke@435 | 321 | return start; |
duke@435 | 322 | } |
duke@435 | 323 | |
duke@435 | 324 | |
duke@435 | 325 | //------------------------------------------------------------------------------------------------------------------------ |
duke@435 | 326 | // Return point for a Java call if there's an exception thrown in Java code. |
duke@435 | 327 | // The exception is caught and transformed into a pending exception stored in |
duke@435 | 328 | // JavaThread that can be tested from within the VM. |
duke@435 | 329 | // |
duke@435 | 330 | // Note: Usually the parameters are removed by the callee. In case of an exception |
duke@435 | 331 | // crossing an activation frame boundary, that is not the case if the callee |
duke@435 | 332 | // is compiled code => need to setup the rsp. |
duke@435 | 333 | // |
duke@435 | 334 | // rax,: exception oop |
duke@435 | 335 | |
duke@435 | 336 | address generate_catch_exception() { |
duke@435 | 337 | StubCodeMark mark(this, "StubRoutines", "catch_exception"); |
duke@435 | 338 | const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()! |
duke@435 | 339 | const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()! |
duke@435 | 340 | address start = __ pc(); |
duke@435 | 341 | |
duke@435 | 342 | // get thread directly |
never@739 | 343 | __ movptr(rcx, thread); |
duke@435 | 344 | #ifdef ASSERT |
duke@435 | 345 | // verify that threads correspond |
duke@435 | 346 | { Label L; |
duke@435 | 347 | __ get_thread(rbx); |
never@739 | 348 | __ cmpptr(rbx, rcx); |
duke@435 | 349 | __ jcc(Assembler::equal, L); |
duke@435 | 350 | __ stop("StubRoutines::catch_exception: threads must correspond"); |
duke@435 | 351 | __ bind(L); |
duke@435 | 352 | } |
duke@435 | 353 | #endif |
duke@435 | 354 | // set pending exception |
duke@435 | 355 | __ verify_oop(rax); |
never@739 | 356 | __ movptr(Address(rcx, Thread::pending_exception_offset()), rax ); |
duke@435 | 357 | __ lea(Address(rcx, Thread::exception_file_offset ()), |
duke@435 | 358 | ExternalAddress((address)__FILE__)); |
duke@435 | 359 | __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ ); |
duke@435 | 360 | // complete return to VM |
duke@435 | 361 | assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before"); |
duke@435 | 362 | __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address)); |
duke@435 | 363 | |
duke@435 | 364 | return start; |
duke@435 | 365 | } |
duke@435 | 366 | |
duke@435 | 367 | |
duke@435 | 368 | //------------------------------------------------------------------------------------------------------------------------ |
duke@435 | 369 | // Continuation point for runtime calls returning with a pending exception. |
duke@435 | 370 | // The pending exception check happened in the runtime or native call stub. |
duke@435 | 371 | // The pending exception in Thread is converted into a Java-level exception. |
duke@435 | 372 | // |
duke@435 | 373 | // Contract with Java-level exception handlers: |
twisti@1730 | 374 | // rax: exception |
duke@435 | 375 | // rdx: throwing pc |
duke@435 | 376 | // |
duke@435 | 377 | // NOTE: At entry of this stub, exception-pc must be on stack !! |
duke@435 | 378 | |
duke@435 | 379 | address generate_forward_exception() { |
duke@435 | 380 | StubCodeMark mark(this, "StubRoutines", "forward exception"); |
duke@435 | 381 | address start = __ pc(); |
twisti@1730 | 382 | const Register thread = rcx; |
twisti@1730 | 383 | |
twisti@1730 | 384 | // other registers used in this stub |
twisti@1730 | 385 | const Register exception_oop = rax; |
twisti@1730 | 386 | const Register handler_addr = rbx; |
twisti@1730 | 387 | const Register exception_pc = rdx; |
duke@435 | 388 | |
duke@435 | 389 | // Upon entry, the sp points to the return address returning into Java |
duke@435 | 390 | // (interpreted or compiled) code; i.e., the return address becomes the |
duke@435 | 391 | // throwing pc. |
duke@435 | 392 | // |
duke@435 | 393 | // Arguments pushed before the runtime call are still on the stack but |
duke@435 | 394 | // the exception handler will reset the stack pointer -> ignore them. |
duke@435 | 395 | // A potential result in registers can be ignored as well. |
duke@435 | 396 | |
duke@435 | 397 | #ifdef ASSERT |
duke@435 | 398 | // make sure this code is only executed if there is a pending exception |
duke@435 | 399 | { Label L; |
twisti@1730 | 400 | __ get_thread(thread); |
twisti@1730 | 401 | __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
duke@435 | 402 | __ jcc(Assembler::notEqual, L); |
duke@435 | 403 | __ stop("StubRoutines::forward exception: no pending exception (1)"); |
duke@435 | 404 | __ bind(L); |
duke@435 | 405 | } |
duke@435 | 406 | #endif |
duke@435 | 407 | |
duke@435 | 408 | // compute exception handler into rbx, |
twisti@1730 | 409 | __ get_thread(thread); |
twisti@1730 | 410 | __ movptr(exception_pc, Address(rsp, 0)); |
duke@435 | 411 | BLOCK_COMMENT("call exception_handler_for_return_address"); |
twisti@1730 | 412 | __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc); |
twisti@1730 | 413 | __ mov(handler_addr, rax); |
duke@435 | 414 | |
twisti@1730 | 415 | // setup rax & rdx, remove return address & clear pending exception |
twisti@1730 | 416 | __ get_thread(thread); |
twisti@1730 | 417 | __ pop(exception_pc); |
twisti@1730 | 418 | __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset())); |
twisti@1730 | 419 | __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD); |
duke@435 | 420 | |
duke@435 | 421 | #ifdef ASSERT |
duke@435 | 422 | // make sure exception is set |
duke@435 | 423 | { Label L; |
twisti@1730 | 424 | __ testptr(exception_oop, exception_oop); |
duke@435 | 425 | __ jcc(Assembler::notEqual, L); |
duke@435 | 426 | __ stop("StubRoutines::forward exception: no pending exception (2)"); |
duke@435 | 427 | __ bind(L); |
duke@435 | 428 | } |
duke@435 | 429 | #endif |
duke@435 | 430 | |
twisti@1730 | 431 | // Verify that there is really a valid exception in RAX. |
twisti@1730 | 432 | __ verify_oop(exception_oop); |
twisti@1730 | 433 | |
duke@435 | 434 | // continue at exception handler (return address removed) |
twisti@1730 | 435 | // rax: exception |
twisti@1730 | 436 | // rbx: exception handler |
duke@435 | 437 | // rdx: throwing pc |
twisti@1730 | 438 | __ jmp(handler_addr); |
duke@435 | 439 | |
duke@435 | 440 | return start; |
duke@435 | 441 | } |
duke@435 | 442 | |
duke@435 | 443 | |
duke@435 | 444 | //---------------------------------------------------------------------------------------------------- |
duke@435 | 445 | // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest) |
duke@435 | 446 | // |
duke@435 | 447 | // xchg exists as far back as 8086, lock needed for MP only |
duke@435 | 448 | // Stack layout immediately after call: |
duke@435 | 449 | // |
duke@435 | 450 | // 0 [ret addr ] <--- rsp |
duke@435 | 451 | // 1 [ ex ] |
duke@435 | 452 | // 2 [ dest ] |
duke@435 | 453 | // |
duke@435 | 454 | // Result: *dest <- ex, return (old *dest) |
duke@435 | 455 | // |
duke@435 | 456 | // Note: win32 does not currently use this code |
duke@435 | 457 | |
duke@435 | 458 | address generate_atomic_xchg() { |
duke@435 | 459 | StubCodeMark mark(this, "StubRoutines", "atomic_xchg"); |
duke@435 | 460 | address start = __ pc(); |
duke@435 | 461 | |
never@739 | 462 | __ push(rdx); |
duke@435 | 463 | Address exchange(rsp, 2 * wordSize); |
duke@435 | 464 | Address dest_addr(rsp, 3 * wordSize); |
duke@435 | 465 | __ movl(rax, exchange); |
never@739 | 466 | __ movptr(rdx, dest_addr); |
never@739 | 467 | __ xchgl(rax, Address(rdx, 0)); |
never@739 | 468 | __ pop(rdx); |
duke@435 | 469 | __ ret(0); |
duke@435 | 470 | |
duke@435 | 471 | return start; |
duke@435 | 472 | } |
duke@435 | 473 | |
duke@435 | 474 | //---------------------------------------------------------------------------------------------------- |
duke@435 | 475 | // Support for void verify_mxcsr() |
duke@435 | 476 | // |
duke@435 | 477 | // This routine is used with -Xcheck:jni to verify that native |
duke@435 | 478 | // JNI code does not return to Java code without restoring the |
duke@435 | 479 | // MXCSR register to our expected state. |
duke@435 | 480 | |
duke@435 | 481 | |
duke@435 | 482 | address generate_verify_mxcsr() { |
duke@435 | 483 | StubCodeMark mark(this, "StubRoutines", "verify_mxcsr"); |
duke@435 | 484 | address start = __ pc(); |
duke@435 | 485 | |
duke@435 | 486 | const Address mxcsr_save(rsp, 0); |
duke@435 | 487 | |
duke@435 | 488 | if (CheckJNICalls && UseSSE > 0 ) { |
duke@435 | 489 | Label ok_ret; |
duke@435 | 490 | ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); |
never@739 | 491 | __ push(rax); |
never@739 | 492 | __ subptr(rsp, wordSize); // allocate a temp location |
duke@435 | 493 | __ stmxcsr(mxcsr_save); |
duke@435 | 494 | __ movl(rax, mxcsr_save); |
duke@435 | 495 | __ andl(rax, MXCSR_MASK); |
duke@435 | 496 | __ cmp32(rax, mxcsr_std); |
duke@435 | 497 | __ jcc(Assembler::equal, ok_ret); |
duke@435 | 498 | |
duke@435 | 499 | __ warn("MXCSR changed by native JNI code."); |
duke@435 | 500 | |
duke@435 | 501 | __ ldmxcsr(mxcsr_std); |
duke@435 | 502 | |
duke@435 | 503 | __ bind(ok_ret); |
never@739 | 504 | __ addptr(rsp, wordSize); |
never@739 | 505 | __ pop(rax); |
duke@435 | 506 | } |
duke@435 | 507 | |
duke@435 | 508 | __ ret(0); |
duke@435 | 509 | |
duke@435 | 510 | return start; |
duke@435 | 511 | } |
duke@435 | 512 | |
duke@435 | 513 | |
duke@435 | 514 | //--------------------------------------------------------------------------- |
duke@435 | 515 | // Support for void verify_fpu_cntrl_wrd() |
duke@435 | 516 | // |
duke@435 | 517 | // This routine is used with -Xcheck:jni to verify that native |
duke@435 | 518 | // JNI code does not return to Java code without restoring the |
duke@435 | 519 | // FP control word to our expected state. |
duke@435 | 520 | |
duke@435 | 521 | address generate_verify_fpu_cntrl_wrd() { |
duke@435 | 522 | StubCodeMark mark(this, "StubRoutines", "verify_spcw"); |
duke@435 | 523 | address start = __ pc(); |
duke@435 | 524 | |
duke@435 | 525 | const Address fpu_cntrl_wrd_save(rsp, 0); |
duke@435 | 526 | |
duke@435 | 527 | if (CheckJNICalls) { |
duke@435 | 528 | Label ok_ret; |
never@739 | 529 | __ push(rax); |
never@739 | 530 | __ subptr(rsp, wordSize); // allocate a temp location |
duke@435 | 531 | __ fnstcw(fpu_cntrl_wrd_save); |
duke@435 | 532 | __ movl(rax, fpu_cntrl_wrd_save); |
duke@435 | 533 | __ andl(rax, FPU_CNTRL_WRD_MASK); |
duke@435 | 534 | ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std()); |
duke@435 | 535 | __ cmp32(rax, fpu_std); |
duke@435 | 536 | __ jcc(Assembler::equal, ok_ret); |
duke@435 | 537 | |
duke@435 | 538 | __ warn("Floating point control word changed by native JNI code."); |
duke@435 | 539 | |
duke@435 | 540 | __ fldcw(fpu_std); |
duke@435 | 541 | |
duke@435 | 542 | __ bind(ok_ret); |
never@739 | 543 | __ addptr(rsp, wordSize); |
never@739 | 544 | __ pop(rax); |
duke@435 | 545 | } |
duke@435 | 546 | |
duke@435 | 547 | __ ret(0); |
duke@435 | 548 | |
duke@435 | 549 | return start; |
duke@435 | 550 | } |
duke@435 | 551 | |
duke@435 | 552 | //--------------------------------------------------------------------------- |
duke@435 | 553 | // Wrapper for slow-case handling of double-to-integer conversion |
duke@435 | 554 | // d2i or f2i fast case failed either because it is nan or because |
duke@435 | 555 | // of under/overflow. |
duke@435 | 556 | // Input: FPU TOS: float value |
duke@435 | 557 | // Output: rax, (rdx): integer (long) result |
duke@435 | 558 | |
duke@435 | 559 | address generate_d2i_wrapper(BasicType t, address fcn) { |
duke@435 | 560 | StubCodeMark mark(this, "StubRoutines", "d2i_wrapper"); |
duke@435 | 561 | address start = __ pc(); |
duke@435 | 562 | |
duke@435 | 563 | // Capture info about frame layout |
duke@435 | 564 | enum layout { FPUState_off = 0, |
duke@435 | 565 | rbp_off = FPUStateSizeInWords, |
duke@435 | 566 | rdi_off, |
duke@435 | 567 | rsi_off, |
duke@435 | 568 | rcx_off, |
duke@435 | 569 | rbx_off, |
duke@435 | 570 | saved_argument_off, |
duke@435 | 571 | saved_argument_off2, // 2nd half of double |
duke@435 | 572 | framesize |
duke@435 | 573 | }; |
duke@435 | 574 | |
duke@435 | 575 | assert(FPUStateSizeInWords == 27, "update stack layout"); |
duke@435 | 576 | |
duke@435 | 577 | // Save outgoing argument to stack across push_FPU_state() |
never@739 | 578 | __ subptr(rsp, wordSize * 2); |
duke@435 | 579 | __ fstp_d(Address(rsp, 0)); |
duke@435 | 580 | |
duke@435 | 581 | // Save CPU & FPU state |
never@739 | 582 | __ push(rbx); |
never@739 | 583 | __ push(rcx); |
never@739 | 584 | __ push(rsi); |
never@739 | 585 | __ push(rdi); |
never@739 | 586 | __ push(rbp); |
duke@435 | 587 | __ push_FPU_state(); |
duke@435 | 588 | |
duke@435 | 589 | // push_FPU_state() resets the FP top of stack |
duke@435 | 590 | // Load original double into FP top of stack |
duke@435 | 591 | __ fld_d(Address(rsp, saved_argument_off * wordSize)); |
duke@435 | 592 | // Store double into stack as outgoing argument |
never@739 | 593 | __ subptr(rsp, wordSize*2); |
duke@435 | 594 | __ fst_d(Address(rsp, 0)); |
duke@435 | 595 | |
duke@435 | 596 | // Prepare FPU for doing math in C-land |
duke@435 | 597 | __ empty_FPU_stack(); |
duke@435 | 598 | // Call the C code to massage the double. Result in EAX |
duke@435 | 599 | if (t == T_INT) |
duke@435 | 600 | { BLOCK_COMMENT("SharedRuntime::d2i"); } |
duke@435 | 601 | else if (t == T_LONG) |
duke@435 | 602 | { BLOCK_COMMENT("SharedRuntime::d2l"); } |
duke@435 | 603 | __ call_VM_leaf( fcn, 2 ); |
duke@435 | 604 | |
duke@435 | 605 | // Restore CPU & FPU state |
duke@435 | 606 | __ pop_FPU_state(); |
never@739 | 607 | __ pop(rbp); |
never@739 | 608 | __ pop(rdi); |
never@739 | 609 | __ pop(rsi); |
never@739 | 610 | __ pop(rcx); |
never@739 | 611 | __ pop(rbx); |
never@739 | 612 | __ addptr(rsp, wordSize * 2); |
duke@435 | 613 | |
duke@435 | 614 | __ ret(0); |
duke@435 | 615 | |
duke@435 | 616 | return start; |
duke@435 | 617 | } |
duke@435 | 618 | |
duke@435 | 619 | |
duke@435 | 620 | //--------------------------------------------------------------------------- |
duke@435 | 621 | // The following routine generates a subroutine to throw an asynchronous |
duke@435 | 622 | // UnknownError when an unsafe access gets a fault that could not be |
duke@435 | 623 | // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.) |
duke@435 | 624 | address generate_handler_for_unsafe_access() { |
duke@435 | 625 | StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access"); |
duke@435 | 626 | address start = __ pc(); |
duke@435 | 627 | |
never@739 | 628 | __ push(0); // hole for return address-to-be |
never@739 | 629 | __ pusha(); // push registers |
duke@435 | 630 | Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord); |
duke@435 | 631 | BLOCK_COMMENT("call handle_unsafe_access"); |
duke@435 | 632 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access))); |
never@739 | 633 | __ movptr(next_pc, rax); // stuff next address |
never@739 | 634 | __ popa(); |
duke@435 | 635 | __ ret(0); // jump to next address |
duke@435 | 636 | |
duke@435 | 637 | return start; |
duke@435 | 638 | } |
duke@435 | 639 | |
duke@435 | 640 | |
duke@435 | 641 | //---------------------------------------------------------------------------------------------------- |
duke@435 | 642 | // Non-destructive plausibility checks for oops |
duke@435 | 643 | |
duke@435 | 644 | address generate_verify_oop() { |
duke@435 | 645 | StubCodeMark mark(this, "StubRoutines", "verify_oop"); |
duke@435 | 646 | address start = __ pc(); |
duke@435 | 647 | |
duke@435 | 648 | // Incoming arguments on stack after saving rax,: |
duke@435 | 649 | // |
duke@435 | 650 | // [tos ]: saved rdx |
duke@435 | 651 | // [tos + 1]: saved EFLAGS |
duke@435 | 652 | // [tos + 2]: return address |
duke@435 | 653 | // [tos + 3]: char* error message |
duke@435 | 654 | // [tos + 4]: oop object to verify |
duke@435 | 655 | // [tos + 5]: saved rax, - saved by caller and bashed |
duke@435 | 656 | |
duke@435 | 657 | Label exit, error; |
never@739 | 658 | __ pushf(); |
never@739 | 659 | __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr())); |
never@739 | 660 | __ push(rdx); // save rdx |
duke@435 | 661 | // make sure object is 'reasonable' |
never@739 | 662 | __ movptr(rax, Address(rsp, 4 * wordSize)); // get object |
never@739 | 663 | __ testptr(rax, rax); |
duke@435 | 664 | __ jcc(Assembler::zero, exit); // if obj is NULL it is ok |
duke@435 | 665 | |
duke@435 | 666 | // Check if the oop is in the right area of memory |
duke@435 | 667 | const int oop_mask = Universe::verify_oop_mask(); |
duke@435 | 668 | const int oop_bits = Universe::verify_oop_bits(); |
never@739 | 669 | __ mov(rdx, rax); |
never@739 | 670 | __ andptr(rdx, oop_mask); |
never@739 | 671 | __ cmpptr(rdx, oop_bits); |
duke@435 | 672 | __ jcc(Assembler::notZero, error); |
duke@435 | 673 | |
coleenp@4037 | 674 | // make sure klass is 'reasonable', which is not zero. |
never@739 | 675 | __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass |
never@739 | 676 | __ testptr(rax, rax); |
duke@435 | 677 | __ jcc(Assembler::zero, error); // if klass is NULL it is broken |
duke@435 | 678 | |
duke@435 | 679 | // return if everything seems ok |
duke@435 | 680 | __ bind(exit); |
never@739 | 681 | __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back |
never@739 | 682 | __ pop(rdx); // restore rdx |
never@739 | 683 | __ popf(); // restore EFLAGS |
duke@435 | 684 | __ ret(3 * wordSize); // pop arguments |
duke@435 | 685 | |
duke@435 | 686 | // handle errors |
duke@435 | 687 | __ bind(error); |
never@739 | 688 | __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back |
never@739 | 689 | __ pop(rdx); // get saved rdx back |
never@739 | 690 | __ popf(); // get saved EFLAGS off stack -- will be ignored |
never@739 | 691 | __ pusha(); // push registers (eip = return address & msg are already pushed) |
duke@435 | 692 | BLOCK_COMMENT("call MacroAssembler::debug"); |
never@739 | 693 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32))); |
never@739 | 694 | __ popa(); |
duke@435 | 695 | __ ret(3 * wordSize); // pop arguments |
duke@435 | 696 | return start; |
duke@435 | 697 | } |
duke@435 | 698 | |
duke@435 | 699 | // |
duke@435 | 700 | // Generate pre-barrier for array stores |
duke@435 | 701 | // |
duke@435 | 702 | // Input: |
duke@435 | 703 | // start - starting address |
ysr@1280 | 704 | // count - element count |
iveresov@2606 | 705 | void gen_write_ref_array_pre_barrier(Register start, Register count, bool uninitialized_target) { |
duke@435 | 706 | assert_different_registers(start, count); |
duke@435 | 707 | BarrierSet* bs = Universe::heap()->barrier_set(); |
duke@435 | 708 | switch (bs->kind()) { |
duke@435 | 709 | case BarrierSet::G1SATBCT: |
duke@435 | 710 | case BarrierSet::G1SATBCTLogging: |
iveresov@2606 | 711 | // With G1, don't generate the call if we statically know that the target in uninitialized |
iveresov@2606 | 712 | if (!uninitialized_target) { |
iveresov@2606 | 713 | __ pusha(); // push registers |
iveresov@2606 | 714 | __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), |
iveresov@2606 | 715 | start, count); |
iveresov@2606 | 716 | __ popa(); |
iveresov@2606 | 717 | } |
duke@435 | 718 | break; |
duke@435 | 719 | case BarrierSet::CardTableModRef: |
duke@435 | 720 | case BarrierSet::CardTableExtension: |
duke@435 | 721 | case BarrierSet::ModRef: |
duke@435 | 722 | break; |
duke@435 | 723 | default : |
duke@435 | 724 | ShouldNotReachHere(); |
duke@435 | 725 | |
duke@435 | 726 | } |
duke@435 | 727 | } |
duke@435 | 728 | |
duke@435 | 729 | |
duke@435 | 730 | // |
duke@435 | 731 | // Generate a post-barrier for an array store |
duke@435 | 732 | // |
duke@435 | 733 | // start - starting address |
duke@435 | 734 | // count - element count |
duke@435 | 735 | // |
duke@435 | 736 | // The two input registers are overwritten. |
duke@435 | 737 | // |
duke@435 | 738 | void gen_write_ref_array_post_barrier(Register start, Register count) { |
duke@435 | 739 | BarrierSet* bs = Universe::heap()->barrier_set(); |
duke@435 | 740 | assert_different_registers(start, count); |
duke@435 | 741 | switch (bs->kind()) { |
duke@435 | 742 | case BarrierSet::G1SATBCT: |
duke@435 | 743 | case BarrierSet::G1SATBCTLogging: |
duke@435 | 744 | { |
never@739 | 745 | __ pusha(); // push registers |
apetrusenko@1627 | 746 | __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), |
apetrusenko@1627 | 747 | start, count); |
never@739 | 748 | __ popa(); |
duke@435 | 749 | } |
duke@435 | 750 | break; |
duke@435 | 751 | |
duke@435 | 752 | case BarrierSet::CardTableModRef: |
duke@435 | 753 | case BarrierSet::CardTableExtension: |
duke@435 | 754 | { |
duke@435 | 755 | CardTableModRefBS* ct = (CardTableModRefBS*)bs; |
duke@435 | 756 | assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); |
duke@435 | 757 | |
duke@435 | 758 | Label L_loop; |
duke@435 | 759 | const Register end = count; // elements count; end == start+count-1 |
duke@435 | 760 | assert_different_registers(start, end); |
duke@435 | 761 | |
never@739 | 762 | __ lea(end, Address(start, count, Address::times_ptr, -wordSize)); |
never@739 | 763 | __ shrptr(start, CardTableModRefBS::card_shift); |
never@739 | 764 | __ shrptr(end, CardTableModRefBS::card_shift); |
never@739 | 765 | __ subptr(end, start); // end --> count |
duke@435 | 766 | __ BIND(L_loop); |
never@684 | 767 | intptr_t disp = (intptr_t) ct->byte_map_base; |
never@684 | 768 | Address cardtable(start, count, Address::times_1, disp); |
never@684 | 769 | __ movb(cardtable, 0); |
duke@435 | 770 | __ decrement(count); |
duke@435 | 771 | __ jcc(Assembler::greaterEqual, L_loop); |
duke@435 | 772 | } |
duke@435 | 773 | break; |
duke@435 | 774 | case BarrierSet::ModRef: |
duke@435 | 775 | break; |
duke@435 | 776 | default : |
duke@435 | 777 | ShouldNotReachHere(); |
duke@435 | 778 | |
duke@435 | 779 | } |
duke@435 | 780 | } |
duke@435 | 781 | |
kvn@840 | 782 | |
kvn@840 | 783 | // Copy 64 bytes chunks |
kvn@840 | 784 | // |
kvn@840 | 785 | // Inputs: |
kvn@840 | 786 | // from - source array address |
kvn@840 | 787 | // to_from - destination array address - from |
kvn@840 | 788 | // qword_count - 8-bytes element count, negative |
kvn@840 | 789 | // |
kvn@840 | 790 | void xmm_copy_forward(Register from, Register to_from, Register qword_count) { |
kvn@840 | 791 | assert( UseSSE >= 2, "supported cpu only" ); |
kvn@840 | 792 | Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit; |
kvn@840 | 793 | // Copy 64-byte chunks |
kvn@840 | 794 | __ jmpb(L_copy_64_bytes); |
kvn@1800 | 795 | __ align(OptoLoopAlignment); |
kvn@840 | 796 | __ BIND(L_copy_64_bytes_loop); |
kvn@840 | 797 | |
kvn@4411 | 798 | if (UseUnalignedLoadStores) { |
kvn@4411 | 799 | if (UseAVX >= 2) { |
kvn@4411 | 800 | __ vmovdqu(xmm0, Address(from, 0)); |
kvn@4411 | 801 | __ vmovdqu(Address(from, to_from, Address::times_1, 0), xmm0); |
kvn@4411 | 802 | __ vmovdqu(xmm1, Address(from, 32)); |
kvn@4411 | 803 | __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1); |
kvn@4411 | 804 | } else { |
kvn@4411 | 805 | __ movdqu(xmm0, Address(from, 0)); |
kvn@4411 | 806 | __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0); |
kvn@4411 | 807 | __ movdqu(xmm1, Address(from, 16)); |
kvn@4411 | 808 | __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1); |
kvn@4411 | 809 | __ movdqu(xmm2, Address(from, 32)); |
kvn@4411 | 810 | __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2); |
kvn@4411 | 811 | __ movdqu(xmm3, Address(from, 48)); |
kvn@4411 | 812 | __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3); |
kvn@4411 | 813 | } |
kvn@840 | 814 | } else { |
kvn@840 | 815 | __ movq(xmm0, Address(from, 0)); |
kvn@840 | 816 | __ movq(Address(from, to_from, Address::times_1, 0), xmm0); |
kvn@840 | 817 | __ movq(xmm1, Address(from, 8)); |
kvn@840 | 818 | __ movq(Address(from, to_from, Address::times_1, 8), xmm1); |
kvn@840 | 819 | __ movq(xmm2, Address(from, 16)); |
kvn@840 | 820 | __ movq(Address(from, to_from, Address::times_1, 16), xmm2); |
kvn@840 | 821 | __ movq(xmm3, Address(from, 24)); |
kvn@840 | 822 | __ movq(Address(from, to_from, Address::times_1, 24), xmm3); |
kvn@840 | 823 | __ movq(xmm4, Address(from, 32)); |
kvn@840 | 824 | __ movq(Address(from, to_from, Address::times_1, 32), xmm4); |
kvn@840 | 825 | __ movq(xmm5, Address(from, 40)); |
kvn@840 | 826 | __ movq(Address(from, to_from, Address::times_1, 40), xmm5); |
kvn@840 | 827 | __ movq(xmm6, Address(from, 48)); |
kvn@840 | 828 | __ movq(Address(from, to_from, Address::times_1, 48), xmm6); |
kvn@840 | 829 | __ movq(xmm7, Address(from, 56)); |
kvn@840 | 830 | __ movq(Address(from, to_from, Address::times_1, 56), xmm7); |
kvn@840 | 831 | } |
kvn@840 | 832 | |
kvn@840 | 833 | __ addl(from, 64); |
kvn@840 | 834 | __ BIND(L_copy_64_bytes); |
kvn@840 | 835 | __ subl(qword_count, 8); |
kvn@840 | 836 | __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop); |
kvn@4873 | 837 | |
kvn@4873 | 838 | if (UseUnalignedLoadStores && (UseAVX >= 2)) { |
kvn@4873 | 839 | // clean upper bits of YMM registers |
kvn@7816 | 840 | __ vpxor(xmm0, xmm0); |
kvn@7816 | 841 | __ vpxor(xmm1, xmm1); |
kvn@4873 | 842 | } |
kvn@840 | 843 | __ addl(qword_count, 8); |
kvn@840 | 844 | __ jccb(Assembler::zero, L_exit); |
kvn@840 | 845 | // |
kvn@840 | 846 | // length is too short, just copy qwords |
kvn@840 | 847 | // |
kvn@840 | 848 | __ BIND(L_copy_8_bytes); |
kvn@840 | 849 | __ movq(xmm0, Address(from, 0)); |
kvn@840 | 850 | __ movq(Address(from, to_from, Address::times_1), xmm0); |
kvn@840 | 851 | __ addl(from, 8); |
kvn@840 | 852 | __ decrement(qword_count); |
kvn@840 | 853 | __ jcc(Assembler::greater, L_copy_8_bytes); |
kvn@840 | 854 | __ BIND(L_exit); |
kvn@840 | 855 | } |
kvn@840 | 856 | |
duke@435 | 857 | // Copy 64 bytes chunks |
duke@435 | 858 | // |
duke@435 | 859 | // Inputs: |
duke@435 | 860 | // from - source array address |
duke@435 | 861 | // to_from - destination array address - from |
duke@435 | 862 | // qword_count - 8-bytes element count, negative |
duke@435 | 863 | // |
duke@435 | 864 | void mmx_copy_forward(Register from, Register to_from, Register qword_count) { |
kvn@840 | 865 | assert( VM_Version::supports_mmx(), "supported cpu only" ); |
duke@435 | 866 | Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit; |
duke@435 | 867 | // Copy 64-byte chunks |
duke@435 | 868 | __ jmpb(L_copy_64_bytes); |
kvn@1800 | 869 | __ align(OptoLoopAlignment); |
duke@435 | 870 | __ BIND(L_copy_64_bytes_loop); |
duke@435 | 871 | __ movq(mmx0, Address(from, 0)); |
duke@435 | 872 | __ movq(mmx1, Address(from, 8)); |
duke@435 | 873 | __ movq(mmx2, Address(from, 16)); |
duke@435 | 874 | __ movq(Address(from, to_from, Address::times_1, 0), mmx0); |
duke@435 | 875 | __ movq(mmx3, Address(from, 24)); |
duke@435 | 876 | __ movq(Address(from, to_from, Address::times_1, 8), mmx1); |
duke@435 | 877 | __ movq(mmx4, Address(from, 32)); |
duke@435 | 878 | __ movq(Address(from, to_from, Address::times_1, 16), mmx2); |
duke@435 | 879 | __ movq(mmx5, Address(from, 40)); |
duke@435 | 880 | __ movq(Address(from, to_from, Address::times_1, 24), mmx3); |
duke@435 | 881 | __ movq(mmx6, Address(from, 48)); |
duke@435 | 882 | __ movq(Address(from, to_from, Address::times_1, 32), mmx4); |
duke@435 | 883 | __ movq(mmx7, Address(from, 56)); |
duke@435 | 884 | __ movq(Address(from, to_from, Address::times_1, 40), mmx5); |
duke@435 | 885 | __ movq(Address(from, to_from, Address::times_1, 48), mmx6); |
duke@435 | 886 | __ movq(Address(from, to_from, Address::times_1, 56), mmx7); |
never@739 | 887 | __ addptr(from, 64); |
duke@435 | 888 | __ BIND(L_copy_64_bytes); |
duke@435 | 889 | __ subl(qword_count, 8); |
duke@435 | 890 | __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop); |
duke@435 | 891 | __ addl(qword_count, 8); |
duke@435 | 892 | __ jccb(Assembler::zero, L_exit); |
duke@435 | 893 | // |
duke@435 | 894 | // length is too short, just copy qwords |
duke@435 | 895 | // |
duke@435 | 896 | __ BIND(L_copy_8_bytes); |
duke@435 | 897 | __ movq(mmx0, Address(from, 0)); |
duke@435 | 898 | __ movq(Address(from, to_from, Address::times_1), mmx0); |
never@739 | 899 | __ addptr(from, 8); |
duke@435 | 900 | __ decrement(qword_count); |
duke@435 | 901 | __ jcc(Assembler::greater, L_copy_8_bytes); |
duke@435 | 902 | __ BIND(L_exit); |
duke@435 | 903 | __ emms(); |
duke@435 | 904 | } |
duke@435 | 905 | |
duke@435 | 906 | address generate_disjoint_copy(BasicType t, bool aligned, |
duke@435 | 907 | Address::ScaleFactor sf, |
iveresov@2606 | 908 | address* entry, const char *name, |
iveresov@2606 | 909 | bool dest_uninitialized = false) { |
duke@435 | 910 | __ align(CodeEntryAlignment); |
duke@435 | 911 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 912 | address start = __ pc(); |
duke@435 | 913 | |
duke@435 | 914 | Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte; |
duke@435 | 915 | Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes; |
duke@435 | 916 | |
never@739 | 917 | int shift = Address::times_ptr - sf; |
duke@435 | 918 | |
duke@435 | 919 | const Register from = rsi; // source array address |
duke@435 | 920 | const Register to = rdi; // destination array address |
duke@435 | 921 | const Register count = rcx; // elements count |
duke@435 | 922 | const Register to_from = to; // (to - from) |
duke@435 | 923 | const Register saved_to = rdx; // saved destination array address |
duke@435 | 924 | |
duke@435 | 925 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
never@739 | 926 | __ push(rsi); |
never@739 | 927 | __ push(rdi); |
never@739 | 928 | __ movptr(from , Address(rsp, 12+ 4)); |
never@739 | 929 | __ movptr(to , Address(rsp, 12+ 8)); |
duke@435 | 930 | __ movl(count, Address(rsp, 12+ 12)); |
iveresov@2595 | 931 | |
iveresov@2595 | 932 | if (entry != NULL) { |
iveresov@2595 | 933 | *entry = __ pc(); // Entry point from conjoint arraycopy stub. |
iveresov@2595 | 934 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 935 | } |
iveresov@2595 | 936 | |
duke@435 | 937 | if (t == T_OBJECT) { |
duke@435 | 938 | __ testl(count, count); |
duke@435 | 939 | __ jcc(Assembler::zero, L_0_count); |
iveresov@2606 | 940 | gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); |
never@739 | 941 | __ mov(saved_to, to); // save 'to' |
duke@435 | 942 | } |
duke@435 | 943 | |
never@739 | 944 | __ subptr(to, from); // to --> to_from |
duke@435 | 945 | __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element |
duke@435 | 946 | __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp |
kvn@840 | 947 | if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) { |
duke@435 | 948 | // align source address at 4 bytes address boundary |
duke@435 | 949 | if (t == T_BYTE) { |
duke@435 | 950 | // One byte misalignment happens only for byte arrays |
duke@435 | 951 | __ testl(from, 1); |
duke@435 | 952 | __ jccb(Assembler::zero, L_skip_align1); |
duke@435 | 953 | __ movb(rax, Address(from, 0)); |
duke@435 | 954 | __ movb(Address(from, to_from, Address::times_1, 0), rax); |
duke@435 | 955 | __ increment(from); |
duke@435 | 956 | __ decrement(count); |
duke@435 | 957 | __ BIND(L_skip_align1); |
duke@435 | 958 | } |
duke@435 | 959 | // Two bytes misalignment happens only for byte and short (char) arrays |
duke@435 | 960 | __ testl(from, 2); |
duke@435 | 961 | __ jccb(Assembler::zero, L_skip_align2); |
duke@435 | 962 | __ movw(rax, Address(from, 0)); |
duke@435 | 963 | __ movw(Address(from, to_from, Address::times_1, 0), rax); |
never@739 | 964 | __ addptr(from, 2); |
duke@435 | 965 | __ subl(count, 1<<(shift-1)); |
duke@435 | 966 | __ BIND(L_skip_align2); |
duke@435 | 967 | } |
duke@435 | 968 | if (!VM_Version::supports_mmx()) { |
never@739 | 969 | __ mov(rax, count); // save 'count' |
never@739 | 970 | __ shrl(count, shift); // bytes count |
never@739 | 971 | __ addptr(to_from, from);// restore 'to' |
never@739 | 972 | __ rep_mov(); |
never@739 | 973 | __ subptr(to_from, from);// restore 'to_from' |
never@739 | 974 | __ mov(count, rax); // restore 'count' |
duke@435 | 975 | __ jmpb(L_copy_2_bytes); // all dwords were copied |
duke@435 | 976 | } else { |
kvn@840 | 977 | if (!UseUnalignedLoadStores) { |
kvn@840 | 978 | // align to 8 bytes, we know we are 4 byte aligned to start |
kvn@840 | 979 | __ testptr(from, 4); |
kvn@840 | 980 | __ jccb(Assembler::zero, L_copy_64_bytes); |
kvn@840 | 981 | __ movl(rax, Address(from, 0)); |
kvn@840 | 982 | __ movl(Address(from, to_from, Address::times_1, 0), rax); |
kvn@840 | 983 | __ addptr(from, 4); |
kvn@840 | 984 | __ subl(count, 1<<shift); |
kvn@840 | 985 | } |
duke@435 | 986 | __ BIND(L_copy_64_bytes); |
never@739 | 987 | __ mov(rax, count); |
duke@435 | 988 | __ shrl(rax, shift+1); // 8 bytes chunk count |
duke@435 | 989 | // |
duke@435 | 990 | // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop |
duke@435 | 991 | // |
kvn@840 | 992 | if (UseXMMForArrayCopy) { |
kvn@840 | 993 | xmm_copy_forward(from, to_from, rax); |
kvn@840 | 994 | } else { |
kvn@840 | 995 | mmx_copy_forward(from, to_from, rax); |
kvn@840 | 996 | } |
duke@435 | 997 | } |
duke@435 | 998 | // copy tailing dword |
duke@435 | 999 | __ BIND(L_copy_4_bytes); |
duke@435 | 1000 | __ testl(count, 1<<shift); |
duke@435 | 1001 | __ jccb(Assembler::zero, L_copy_2_bytes); |
duke@435 | 1002 | __ movl(rax, Address(from, 0)); |
duke@435 | 1003 | __ movl(Address(from, to_from, Address::times_1, 0), rax); |
duke@435 | 1004 | if (t == T_BYTE || t == T_SHORT) { |
never@739 | 1005 | __ addptr(from, 4); |
duke@435 | 1006 | __ BIND(L_copy_2_bytes); |
duke@435 | 1007 | // copy tailing word |
duke@435 | 1008 | __ testl(count, 1<<(shift-1)); |
duke@435 | 1009 | __ jccb(Assembler::zero, L_copy_byte); |
duke@435 | 1010 | __ movw(rax, Address(from, 0)); |
duke@435 | 1011 | __ movw(Address(from, to_from, Address::times_1, 0), rax); |
duke@435 | 1012 | if (t == T_BYTE) { |
never@739 | 1013 | __ addptr(from, 2); |
duke@435 | 1014 | __ BIND(L_copy_byte); |
duke@435 | 1015 | // copy tailing byte |
duke@435 | 1016 | __ testl(count, 1); |
duke@435 | 1017 | __ jccb(Assembler::zero, L_exit); |
duke@435 | 1018 | __ movb(rax, Address(from, 0)); |
duke@435 | 1019 | __ movb(Address(from, to_from, Address::times_1, 0), rax); |
duke@435 | 1020 | __ BIND(L_exit); |
duke@435 | 1021 | } else { |
duke@435 | 1022 | __ BIND(L_copy_byte); |
duke@435 | 1023 | } |
duke@435 | 1024 | } else { |
duke@435 | 1025 | __ BIND(L_copy_2_bytes); |
duke@435 | 1026 | } |
duke@435 | 1027 | |
duke@435 | 1028 | if (t == T_OBJECT) { |
duke@435 | 1029 | __ movl(count, Address(rsp, 12+12)); // reread 'count' |
never@739 | 1030 | __ mov(to, saved_to); // restore 'to' |
duke@435 | 1031 | gen_write_ref_array_post_barrier(to, count); |
duke@435 | 1032 | __ BIND(L_0_count); |
duke@435 | 1033 | } |
duke@435 | 1034 | inc_copy_counter_np(t); |
never@739 | 1035 | __ pop(rdi); |
never@739 | 1036 | __ pop(rsi); |
duke@435 | 1037 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
never@739 | 1038 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1039 | __ ret(0); |
duke@435 | 1040 | return start; |
duke@435 | 1041 | } |
duke@435 | 1042 | |
duke@435 | 1043 | |
never@2118 | 1044 | address generate_fill(BasicType t, bool aligned, const char *name) { |
never@2118 | 1045 | __ align(CodeEntryAlignment); |
never@2118 | 1046 | StubCodeMark mark(this, "StubRoutines", name); |
never@2118 | 1047 | address start = __ pc(); |
never@2118 | 1048 | |
never@2118 | 1049 | BLOCK_COMMENT("Entry:"); |
never@2118 | 1050 | |
never@2118 | 1051 | const Register to = rdi; // source array address |
never@2118 | 1052 | const Register value = rdx; // value |
never@2118 | 1053 | const Register count = rsi; // elements count |
never@2118 | 1054 | |
never@2118 | 1055 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
never@2118 | 1056 | __ push(rsi); |
never@2118 | 1057 | __ push(rdi); |
never@2118 | 1058 | __ movptr(to , Address(rsp, 12+ 4)); |
never@2118 | 1059 | __ movl(value, Address(rsp, 12+ 8)); |
never@2118 | 1060 | __ movl(count, Address(rsp, 12+ 12)); |
never@2118 | 1061 | |
never@2118 | 1062 | __ generate_fill(t, aligned, to, value, count, rax, xmm0); |
never@2118 | 1063 | |
never@2118 | 1064 | __ pop(rdi); |
never@2118 | 1065 | __ pop(rsi); |
never@2118 | 1066 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
never@2118 | 1067 | __ ret(0); |
never@2118 | 1068 | return start; |
never@2118 | 1069 | } |
never@2118 | 1070 | |
duke@435 | 1071 | address generate_conjoint_copy(BasicType t, bool aligned, |
duke@435 | 1072 | Address::ScaleFactor sf, |
duke@435 | 1073 | address nooverlap_target, |
iveresov@2606 | 1074 | address* entry, const char *name, |
iveresov@2606 | 1075 | bool dest_uninitialized = false) { |
duke@435 | 1076 | __ align(CodeEntryAlignment); |
duke@435 | 1077 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1078 | address start = __ pc(); |
duke@435 | 1079 | |
duke@435 | 1080 | Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte; |
duke@435 | 1081 | Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop; |
duke@435 | 1082 | |
never@739 | 1083 | int shift = Address::times_ptr - sf; |
duke@435 | 1084 | |
duke@435 | 1085 | const Register src = rax; // source array address |
duke@435 | 1086 | const Register dst = rdx; // destination array address |
duke@435 | 1087 | const Register from = rsi; // source array address |
duke@435 | 1088 | const Register to = rdi; // destination array address |
duke@435 | 1089 | const Register count = rcx; // elements count |
duke@435 | 1090 | const Register end = rax; // array end address |
duke@435 | 1091 | |
duke@435 | 1092 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
never@739 | 1093 | __ push(rsi); |
never@739 | 1094 | __ push(rdi); |
never@739 | 1095 | __ movptr(src , Address(rsp, 12+ 4)); // from |
never@739 | 1096 | __ movptr(dst , Address(rsp, 12+ 8)); // to |
never@739 | 1097 | __ movl2ptr(count, Address(rsp, 12+12)); // count |
duke@435 | 1098 | |
duke@435 | 1099 | if (entry != NULL) { |
duke@435 | 1100 | *entry = __ pc(); // Entry point from generic arraycopy stub. |
duke@435 | 1101 | BLOCK_COMMENT("Entry:"); |
duke@435 | 1102 | } |
duke@435 | 1103 | |
iveresov@2595 | 1104 | // nooverlap_target expects arguments in rsi and rdi. |
never@739 | 1105 | __ mov(from, src); |
never@739 | 1106 | __ mov(to , dst); |
duke@435 | 1107 | |
iveresov@2595 | 1108 | // arrays overlap test: dispatch to disjoint stub if necessary. |
duke@435 | 1109 | RuntimeAddress nooverlap(nooverlap_target); |
never@739 | 1110 | __ cmpptr(dst, src); |
never@739 | 1111 | __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size |
duke@435 | 1112 | __ jump_cc(Assembler::belowEqual, nooverlap); |
never@739 | 1113 | __ cmpptr(dst, end); |
duke@435 | 1114 | __ jump_cc(Assembler::aboveEqual, nooverlap); |
duke@435 | 1115 | |
iveresov@2595 | 1116 | if (t == T_OBJECT) { |
iveresov@2595 | 1117 | __ testl(count, count); |
iveresov@2595 | 1118 | __ jcc(Assembler::zero, L_0_count); |
iveresov@2606 | 1119 | gen_write_ref_array_pre_barrier(dst, count, dest_uninitialized); |
iveresov@2595 | 1120 | } |
iveresov@2595 | 1121 | |
duke@435 | 1122 | // copy from high to low |
duke@435 | 1123 | __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element |
duke@435 | 1124 | __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp |
duke@435 | 1125 | if (t == T_BYTE || t == T_SHORT) { |
duke@435 | 1126 | // Align the end of destination array at 4 bytes address boundary |
never@739 | 1127 | __ lea(end, Address(dst, count, sf, 0)); |
duke@435 | 1128 | if (t == T_BYTE) { |
duke@435 | 1129 | // One byte misalignment happens only for byte arrays |
duke@435 | 1130 | __ testl(end, 1); |
duke@435 | 1131 | __ jccb(Assembler::zero, L_skip_align1); |
duke@435 | 1132 | __ decrement(count); |
duke@435 | 1133 | __ movb(rdx, Address(from, count, sf, 0)); |
duke@435 | 1134 | __ movb(Address(to, count, sf, 0), rdx); |
duke@435 | 1135 | __ BIND(L_skip_align1); |
duke@435 | 1136 | } |
duke@435 | 1137 | // Two bytes misalignment happens only for byte and short (char) arrays |
duke@435 | 1138 | __ testl(end, 2); |
duke@435 | 1139 | __ jccb(Assembler::zero, L_skip_align2); |
never@739 | 1140 | __ subptr(count, 1<<(shift-1)); |
duke@435 | 1141 | __ movw(rdx, Address(from, count, sf, 0)); |
duke@435 | 1142 | __ movw(Address(to, count, sf, 0), rdx); |
duke@435 | 1143 | __ BIND(L_skip_align2); |
duke@435 | 1144 | __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element |
duke@435 | 1145 | __ jcc(Assembler::below, L_copy_4_bytes); |
duke@435 | 1146 | } |
duke@435 | 1147 | |
duke@435 | 1148 | if (!VM_Version::supports_mmx()) { |
duke@435 | 1149 | __ std(); |
never@739 | 1150 | __ mov(rax, count); // Save 'count' |
never@739 | 1151 | __ mov(rdx, to); // Save 'to' |
never@739 | 1152 | __ lea(rsi, Address(from, count, sf, -4)); |
never@739 | 1153 | __ lea(rdi, Address(to , count, sf, -4)); |
never@739 | 1154 | __ shrptr(count, shift); // bytes count |
never@739 | 1155 | __ rep_mov(); |
duke@435 | 1156 | __ cld(); |
never@739 | 1157 | __ mov(count, rax); // restore 'count' |
duke@435 | 1158 | __ andl(count, (1<<shift)-1); // mask the number of rest elements |
never@739 | 1159 | __ movptr(from, Address(rsp, 12+4)); // reread 'from' |
never@739 | 1160 | __ mov(to, rdx); // restore 'to' |
duke@435 | 1161 | __ jmpb(L_copy_2_bytes); // all dword were copied |
duke@435 | 1162 | } else { |
duke@435 | 1163 | // Align to 8 bytes the end of array. It is aligned to 4 bytes already. |
never@739 | 1164 | __ testptr(end, 4); |
duke@435 | 1165 | __ jccb(Assembler::zero, L_copy_8_bytes); |
duke@435 | 1166 | __ subl(count, 1<<shift); |
duke@435 | 1167 | __ movl(rdx, Address(from, count, sf, 0)); |
duke@435 | 1168 | __ movl(Address(to, count, sf, 0), rdx); |
duke@435 | 1169 | __ jmpb(L_copy_8_bytes); |
duke@435 | 1170 | |
kvn@1800 | 1171 | __ align(OptoLoopAlignment); |
duke@435 | 1172 | // Move 8 bytes |
duke@435 | 1173 | __ BIND(L_copy_8_bytes_loop); |
kvn@840 | 1174 | if (UseXMMForArrayCopy) { |
kvn@840 | 1175 | __ movq(xmm0, Address(from, count, sf, 0)); |
kvn@840 | 1176 | __ movq(Address(to, count, sf, 0), xmm0); |
kvn@840 | 1177 | } else { |
kvn@840 | 1178 | __ movq(mmx0, Address(from, count, sf, 0)); |
kvn@840 | 1179 | __ movq(Address(to, count, sf, 0), mmx0); |
kvn@840 | 1180 | } |
duke@435 | 1181 | __ BIND(L_copy_8_bytes); |
duke@435 | 1182 | __ subl(count, 2<<shift); |
duke@435 | 1183 | __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); |
duke@435 | 1184 | __ addl(count, 2<<shift); |
kvn@840 | 1185 | if (!UseXMMForArrayCopy) { |
kvn@840 | 1186 | __ emms(); |
kvn@840 | 1187 | } |
duke@435 | 1188 | } |
duke@435 | 1189 | __ BIND(L_copy_4_bytes); |
duke@435 | 1190 | // copy prefix qword |
duke@435 | 1191 | __ testl(count, 1<<shift); |
duke@435 | 1192 | __ jccb(Assembler::zero, L_copy_2_bytes); |
duke@435 | 1193 | __ movl(rdx, Address(from, count, sf, -4)); |
duke@435 | 1194 | __ movl(Address(to, count, sf, -4), rdx); |
duke@435 | 1195 | |
duke@435 | 1196 | if (t == T_BYTE || t == T_SHORT) { |
duke@435 | 1197 | __ subl(count, (1<<shift)); |
duke@435 | 1198 | __ BIND(L_copy_2_bytes); |
duke@435 | 1199 | // copy prefix dword |
duke@435 | 1200 | __ testl(count, 1<<(shift-1)); |
duke@435 | 1201 | __ jccb(Assembler::zero, L_copy_byte); |
duke@435 | 1202 | __ movw(rdx, Address(from, count, sf, -2)); |
duke@435 | 1203 | __ movw(Address(to, count, sf, -2), rdx); |
duke@435 | 1204 | if (t == T_BYTE) { |
duke@435 | 1205 | __ subl(count, 1<<(shift-1)); |
duke@435 | 1206 | __ BIND(L_copy_byte); |
duke@435 | 1207 | // copy prefix byte |
duke@435 | 1208 | __ testl(count, 1); |
duke@435 | 1209 | __ jccb(Assembler::zero, L_exit); |
duke@435 | 1210 | __ movb(rdx, Address(from, 0)); |
duke@435 | 1211 | __ movb(Address(to, 0), rdx); |
duke@435 | 1212 | __ BIND(L_exit); |
duke@435 | 1213 | } else { |
duke@435 | 1214 | __ BIND(L_copy_byte); |
duke@435 | 1215 | } |
duke@435 | 1216 | } else { |
duke@435 | 1217 | __ BIND(L_copy_2_bytes); |
duke@435 | 1218 | } |
duke@435 | 1219 | if (t == T_OBJECT) { |
never@739 | 1220 | __ movl2ptr(count, Address(rsp, 12+12)); // reread count |
duke@435 | 1221 | gen_write_ref_array_post_barrier(to, count); |
duke@435 | 1222 | __ BIND(L_0_count); |
duke@435 | 1223 | } |
duke@435 | 1224 | inc_copy_counter_np(t); |
never@739 | 1225 | __ pop(rdi); |
never@739 | 1226 | __ pop(rsi); |
duke@435 | 1227 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
never@739 | 1228 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1229 | __ ret(0); |
duke@435 | 1230 | return start; |
duke@435 | 1231 | } |
duke@435 | 1232 | |
duke@435 | 1233 | |
duke@435 | 1234 | address generate_disjoint_long_copy(address* entry, const char *name) { |
duke@435 | 1235 | __ align(CodeEntryAlignment); |
duke@435 | 1236 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1237 | address start = __ pc(); |
duke@435 | 1238 | |
duke@435 | 1239 | Label L_copy_8_bytes, L_copy_8_bytes_loop; |
duke@435 | 1240 | const Register from = rax; // source array address |
duke@435 | 1241 | const Register to = rdx; // destination array address |
duke@435 | 1242 | const Register count = rcx; // elements count |
duke@435 | 1243 | const Register to_from = rdx; // (to - from) |
duke@435 | 1244 | |
duke@435 | 1245 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
never@739 | 1246 | __ movptr(from , Address(rsp, 8+0)); // from |
never@739 | 1247 | __ movptr(to , Address(rsp, 8+4)); // to |
never@739 | 1248 | __ movl2ptr(count, Address(rsp, 8+8)); // count |
duke@435 | 1249 | |
duke@435 | 1250 | *entry = __ pc(); // Entry point from conjoint arraycopy stub. |
duke@435 | 1251 | BLOCK_COMMENT("Entry:"); |
duke@435 | 1252 | |
never@739 | 1253 | __ subptr(to, from); // to --> to_from |
duke@435 | 1254 | if (VM_Version::supports_mmx()) { |
kvn@840 | 1255 | if (UseXMMForArrayCopy) { |
kvn@840 | 1256 | xmm_copy_forward(from, to_from, count); |
kvn@840 | 1257 | } else { |
kvn@840 | 1258 | mmx_copy_forward(from, to_from, count); |
kvn@840 | 1259 | } |
duke@435 | 1260 | } else { |
duke@435 | 1261 | __ jmpb(L_copy_8_bytes); |
kvn@1800 | 1262 | __ align(OptoLoopAlignment); |
duke@435 | 1263 | __ BIND(L_copy_8_bytes_loop); |
duke@435 | 1264 | __ fild_d(Address(from, 0)); |
duke@435 | 1265 | __ fistp_d(Address(from, to_from, Address::times_1)); |
never@739 | 1266 | __ addptr(from, 8); |
duke@435 | 1267 | __ BIND(L_copy_8_bytes); |
duke@435 | 1268 | __ decrement(count); |
duke@435 | 1269 | __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); |
duke@435 | 1270 | } |
duke@435 | 1271 | inc_copy_counter_np(T_LONG); |
duke@435 | 1272 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
never@739 | 1273 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1274 | __ ret(0); |
duke@435 | 1275 | return start; |
duke@435 | 1276 | } |
duke@435 | 1277 | |
duke@435 | 1278 | address generate_conjoint_long_copy(address nooverlap_target, |
duke@435 | 1279 | address* entry, const char *name) { |
duke@435 | 1280 | __ align(CodeEntryAlignment); |
duke@435 | 1281 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1282 | address start = __ pc(); |
duke@435 | 1283 | |
duke@435 | 1284 | Label L_copy_8_bytes, L_copy_8_bytes_loop; |
duke@435 | 1285 | const Register from = rax; // source array address |
duke@435 | 1286 | const Register to = rdx; // destination array address |
duke@435 | 1287 | const Register count = rcx; // elements count |
duke@435 | 1288 | const Register end_from = rax; // source array end address |
duke@435 | 1289 | |
duke@435 | 1290 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
never@739 | 1291 | __ movptr(from , Address(rsp, 8+0)); // from |
never@739 | 1292 | __ movptr(to , Address(rsp, 8+4)); // to |
never@739 | 1293 | __ movl2ptr(count, Address(rsp, 8+8)); // count |
duke@435 | 1294 | |
duke@435 | 1295 | *entry = __ pc(); // Entry point from generic arraycopy stub. |
duke@435 | 1296 | BLOCK_COMMENT("Entry:"); |
duke@435 | 1297 | |
duke@435 | 1298 | // arrays overlap test |
never@739 | 1299 | __ cmpptr(to, from); |
duke@435 | 1300 | RuntimeAddress nooverlap(nooverlap_target); |
duke@435 | 1301 | __ jump_cc(Assembler::belowEqual, nooverlap); |
never@739 | 1302 | __ lea(end_from, Address(from, count, Address::times_8, 0)); |
never@739 | 1303 | __ cmpptr(to, end_from); |
never@739 | 1304 | __ movptr(from, Address(rsp, 8)); // from |
duke@435 | 1305 | __ jump_cc(Assembler::aboveEqual, nooverlap); |
duke@435 | 1306 | |
duke@435 | 1307 | __ jmpb(L_copy_8_bytes); |
duke@435 | 1308 | |
kvn@1800 | 1309 | __ align(OptoLoopAlignment); |
duke@435 | 1310 | __ BIND(L_copy_8_bytes_loop); |
duke@435 | 1311 | if (VM_Version::supports_mmx()) { |
kvn@840 | 1312 | if (UseXMMForArrayCopy) { |
kvn@840 | 1313 | __ movq(xmm0, Address(from, count, Address::times_8)); |
kvn@840 | 1314 | __ movq(Address(to, count, Address::times_8), xmm0); |
kvn@840 | 1315 | } else { |
kvn@840 | 1316 | __ movq(mmx0, Address(from, count, Address::times_8)); |
kvn@840 | 1317 | __ movq(Address(to, count, Address::times_8), mmx0); |
kvn@840 | 1318 | } |
duke@435 | 1319 | } else { |
duke@435 | 1320 | __ fild_d(Address(from, count, Address::times_8)); |
duke@435 | 1321 | __ fistp_d(Address(to, count, Address::times_8)); |
duke@435 | 1322 | } |
duke@435 | 1323 | __ BIND(L_copy_8_bytes); |
duke@435 | 1324 | __ decrement(count); |
duke@435 | 1325 | __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); |
duke@435 | 1326 | |
kvn@840 | 1327 | if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) { |
duke@435 | 1328 | __ emms(); |
duke@435 | 1329 | } |
duke@435 | 1330 | inc_copy_counter_np(T_LONG); |
duke@435 | 1331 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
never@739 | 1332 | __ xorptr(rax, rax); // return 0 |
duke@435 | 1333 | __ ret(0); |
duke@435 | 1334 | return start; |
duke@435 | 1335 | } |
duke@435 | 1336 | |
duke@435 | 1337 | |
duke@435 | 1338 | // Helper for generating a dynamic type check. |
duke@435 | 1339 | // The sub_klass must be one of {rbx, rdx, rsi}. |
duke@435 | 1340 | // The temp is killed. |
duke@435 | 1341 | void generate_type_check(Register sub_klass, |
duke@435 | 1342 | Address& super_check_offset_addr, |
duke@435 | 1343 | Address& super_klass_addr, |
duke@435 | 1344 | Register temp, |
jrose@1079 | 1345 | Label* L_success, Label* L_failure) { |
duke@435 | 1346 | BLOCK_COMMENT("type_check:"); |
duke@435 | 1347 | |
duke@435 | 1348 | Label L_fallthrough; |
jrose@1079 | 1349 | #define LOCAL_JCC(assembler_con, label_ptr) \ |
jrose@1079 | 1350 | if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \ |
jrose@1079 | 1351 | else __ jcc(assembler_con, L_fallthrough) /*omit semi*/ |
duke@435 | 1352 | |
jrose@1079 | 1353 | // The following is a strange variation of the fast path which requires |
jrose@1079 | 1354 | // one less register, because needed values are on the argument stack. |
jrose@1079 | 1355 | // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp, |
jrose@1079 | 1356 | // L_success, L_failure, NULL); |
duke@435 | 1357 | assert_different_registers(sub_klass, temp); |
duke@435 | 1358 | |
stefank@3391 | 1359 | int sc_offset = in_bytes(Klass::secondary_super_cache_offset()); |
duke@435 | 1360 | |
duke@435 | 1361 | // if the pointers are equal, we are done (e.g., String[] elements) |
never@739 | 1362 | __ cmpptr(sub_klass, super_klass_addr); |
jrose@1079 | 1363 | LOCAL_JCC(Assembler::equal, L_success); |
duke@435 | 1364 | |
duke@435 | 1365 | // check the supertype display: |
never@739 | 1366 | __ movl2ptr(temp, super_check_offset_addr); |
duke@435 | 1367 | Address super_check_addr(sub_klass, temp, Address::times_1, 0); |
never@739 | 1368 | __ movptr(temp, super_check_addr); // load displayed supertype |
never@739 | 1369 | __ cmpptr(temp, super_klass_addr); // test the super type |
jrose@1079 | 1370 | LOCAL_JCC(Assembler::equal, L_success); |
duke@435 | 1371 | |
duke@435 | 1372 | // if it was a primary super, we can just fail immediately |
duke@435 | 1373 | __ cmpl(super_check_offset_addr, sc_offset); |
jrose@1079 | 1374 | LOCAL_JCC(Assembler::notEqual, L_failure); |
duke@435 | 1375 | |
jrose@1079 | 1376 | // The repne_scan instruction uses fixed registers, which will get spilled. |
jrose@1079 | 1377 | // We happen to know this works best when super_klass is in rax. |
jrose@1079 | 1378 | Register super_klass = temp; |
jrose@1079 | 1379 | __ movptr(super_klass, super_klass_addr); |
jrose@1079 | 1380 | __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, |
jrose@1079 | 1381 | L_success, L_failure); |
duke@435 | 1382 | |
jrose@1079 | 1383 | __ bind(L_fallthrough); |
duke@435 | 1384 | |
jrose@1079 | 1385 | if (L_success == NULL) { BLOCK_COMMENT("L_success:"); } |
jrose@1079 | 1386 | if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); } |
duke@435 | 1387 | |
jrose@1079 | 1388 | #undef LOCAL_JCC |
duke@435 | 1389 | } |
duke@435 | 1390 | |
duke@435 | 1391 | // |
duke@435 | 1392 | // Generate checkcasting array copy stub |
duke@435 | 1393 | // |
duke@435 | 1394 | // Input: |
duke@435 | 1395 | // 4(rsp) - source array address |
duke@435 | 1396 | // 8(rsp) - destination array address |
duke@435 | 1397 | // 12(rsp) - element count, can be zero |
duke@435 | 1398 | // 16(rsp) - size_t ckoff (super_check_offset) |
duke@435 | 1399 | // 20(rsp) - oop ckval (super_klass) |
duke@435 | 1400 | // |
duke@435 | 1401 | // Output: |
duke@435 | 1402 | // rax, == 0 - success |
duke@435 | 1403 | // rax, == -1^K - failure, where K is partial transfer count |
duke@435 | 1404 | // |
iveresov@2606 | 1405 | address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) { |
duke@435 | 1406 | __ align(CodeEntryAlignment); |
duke@435 | 1407 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1408 | address start = __ pc(); |
duke@435 | 1409 | |
duke@435 | 1410 | Label L_load_element, L_store_element, L_do_card_marks, L_done; |
duke@435 | 1411 | |
duke@435 | 1412 | // register use: |
duke@435 | 1413 | // rax, rdx, rcx -- loop control (end_from, end_to, count) |
duke@435 | 1414 | // rdi, rsi -- element access (oop, klass) |
duke@435 | 1415 | // rbx, -- temp |
duke@435 | 1416 | const Register from = rax; // source array address |
duke@435 | 1417 | const Register to = rdx; // destination array address |
duke@435 | 1418 | const Register length = rcx; // elements count |
duke@435 | 1419 | const Register elem = rdi; // each oop copied |
duke@435 | 1420 | const Register elem_klass = rsi; // each elem._klass (sub_klass) |
duke@435 | 1421 | const Register temp = rbx; // lone remaining temp |
duke@435 | 1422 | |
duke@435 | 1423 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1424 | |
never@739 | 1425 | __ push(rsi); |
never@739 | 1426 | __ push(rdi); |
never@739 | 1427 | __ push(rbx); |
duke@435 | 1428 | |
duke@435 | 1429 | Address from_arg(rsp, 16+ 4); // from |
duke@435 | 1430 | Address to_arg(rsp, 16+ 8); // to |
duke@435 | 1431 | Address length_arg(rsp, 16+12); // elements count |
duke@435 | 1432 | Address ckoff_arg(rsp, 16+16); // super_check_offset |
duke@435 | 1433 | Address ckval_arg(rsp, 16+20); // super_klass |
duke@435 | 1434 | |
duke@435 | 1435 | // Load up: |
never@739 | 1436 | __ movptr(from, from_arg); |
never@739 | 1437 | __ movptr(to, to_arg); |
never@739 | 1438 | __ movl2ptr(length, length_arg); |
duke@435 | 1439 | |
iveresov@2595 | 1440 | if (entry != NULL) { |
iveresov@2595 | 1441 | *entry = __ pc(); // Entry point from generic arraycopy stub. |
iveresov@2595 | 1442 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1443 | } |
duke@435 | 1444 | |
duke@435 | 1445 | //--------------------------------------------------------------- |
duke@435 | 1446 | // Assembler stub will be used for this call to arraycopy |
duke@435 | 1447 | // if the two arrays are subtypes of Object[] but the |
duke@435 | 1448 | // destination array type is not equal to or a supertype |
duke@435 | 1449 | // of the source type. Each element must be separately |
duke@435 | 1450 | // checked. |
duke@435 | 1451 | |
duke@435 | 1452 | // Loop-invariant addresses. They are exclusive end pointers. |
never@739 | 1453 | Address end_from_addr(from, length, Address::times_ptr, 0); |
never@739 | 1454 | Address end_to_addr(to, length, Address::times_ptr, 0); |
duke@435 | 1455 | |
duke@435 | 1456 | Register end_from = from; // re-use |
duke@435 | 1457 | Register end_to = to; // re-use |
duke@435 | 1458 | Register count = length; // re-use |
duke@435 | 1459 | |
duke@435 | 1460 | // Loop-variant addresses. They assume post-incremented count < 0. |
never@739 | 1461 | Address from_element_addr(end_from, count, Address::times_ptr, 0); |
never@739 | 1462 | Address to_element_addr(end_to, count, Address::times_ptr, 0); |
duke@435 | 1463 | Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes()); |
duke@435 | 1464 | |
duke@435 | 1465 | // Copy from low to high addresses, indexed from the end of each array. |
iveresov@2606 | 1466 | gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); |
never@739 | 1467 | __ lea(end_from, end_from_addr); |
never@739 | 1468 | __ lea(end_to, end_to_addr); |
duke@435 | 1469 | assert(length == count, ""); // else fix next line: |
never@739 | 1470 | __ negptr(count); // negate and test the length |
duke@435 | 1471 | __ jccb(Assembler::notZero, L_load_element); |
duke@435 | 1472 | |
duke@435 | 1473 | // Empty array: Nothing to do. |
never@739 | 1474 | __ xorptr(rax, rax); // return 0 on (trivial) success |
duke@435 | 1475 | __ jmp(L_done); |
duke@435 | 1476 | |
duke@435 | 1477 | // ======== begin loop ======== |
duke@435 | 1478 | // (Loop is rotated; its entry is L_load_element.) |
duke@435 | 1479 | // Loop control: |
duke@435 | 1480 | // for (count = -count; count != 0; count++) |
duke@435 | 1481 | // Base pointers src, dst are biased by 8*count,to last element. |
kvn@1800 | 1482 | __ align(OptoLoopAlignment); |
duke@435 | 1483 | |
duke@435 | 1484 | __ BIND(L_store_element); |
never@739 | 1485 | __ movptr(to_element_addr, elem); // store the oop |
duke@435 | 1486 | __ increment(count); // increment the count toward zero |
duke@435 | 1487 | __ jccb(Assembler::zero, L_do_card_marks); |
duke@435 | 1488 | |
duke@435 | 1489 | // ======== loop entry is here ======== |
duke@435 | 1490 | __ BIND(L_load_element); |
never@739 | 1491 | __ movptr(elem, from_element_addr); // load the oop |
never@739 | 1492 | __ testptr(elem, elem); |
duke@435 | 1493 | __ jccb(Assembler::zero, L_store_element); |
duke@435 | 1494 | |
duke@435 | 1495 | // (Could do a trick here: Remember last successful non-null |
duke@435 | 1496 | // element stored and make a quick oop equality check on it.) |
duke@435 | 1497 | |
never@739 | 1498 | __ movptr(elem_klass, elem_klass_addr); // query the object klass |
duke@435 | 1499 | generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp, |
duke@435 | 1500 | &L_store_element, NULL); |
kvn@5156 | 1501 | // (On fall-through, we have failed the element type check.) |
duke@435 | 1502 | // ======== end loop ======== |
duke@435 | 1503 | |
duke@435 | 1504 | // It was a real error; we must depend on the caller to finish the job. |
rasbold@454 | 1505 | // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops. |
rasbold@454 | 1506 | // Emit GC store barriers for the oops we have copied (length_arg + count), |
duke@435 | 1507 | // and report their number to the caller. |
kvn@5156 | 1508 | assert_different_registers(to, count, rax); |
kvn@5156 | 1509 | Label L_post_barrier; |
duke@435 | 1510 | __ addl(count, length_arg); // transfers = (length - remaining) |
never@739 | 1511 | __ movl2ptr(rax, count); // save the value |
kvn@5156 | 1512 | __ notptr(rax); // report (-1^K) to caller (does not affect flags) |
kvn@5156 | 1513 | __ jccb(Assembler::notZero, L_post_barrier); |
kvn@5156 | 1514 | __ jmp(L_done); // K == 0, nothing was copied, skip post barrier |
duke@435 | 1515 | |
duke@435 | 1516 | // Come here on success only. |
duke@435 | 1517 | __ BIND(L_do_card_marks); |
kvn@5156 | 1518 | __ xorptr(rax, rax); // return 0 on success |
never@739 | 1519 | __ movl2ptr(count, length_arg); |
kvn@5156 | 1520 | |
kvn@5156 | 1521 | __ BIND(L_post_barrier); |
kvn@5156 | 1522 | __ movptr(to, to_arg); // reload |
duke@435 | 1523 | gen_write_ref_array_post_barrier(to, count); |
duke@435 | 1524 | |
duke@435 | 1525 | // Common exit point (success or failure). |
duke@435 | 1526 | __ BIND(L_done); |
never@739 | 1527 | __ pop(rbx); |
never@739 | 1528 | __ pop(rdi); |
never@739 | 1529 | __ pop(rsi); |
duke@435 | 1530 | inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); |
duke@435 | 1531 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1532 | __ ret(0); |
duke@435 | 1533 | |
duke@435 | 1534 | return start; |
duke@435 | 1535 | } |
duke@435 | 1536 | |
duke@435 | 1537 | // |
duke@435 | 1538 | // Generate 'unsafe' array copy stub |
duke@435 | 1539 | // Though just as safe as the other stubs, it takes an unscaled |
duke@435 | 1540 | // size_t argument instead of an element count. |
duke@435 | 1541 | // |
duke@435 | 1542 | // Input: |
duke@435 | 1543 | // 4(rsp) - source array address |
duke@435 | 1544 | // 8(rsp) - destination array address |
duke@435 | 1545 | // 12(rsp) - byte count, can be zero |
duke@435 | 1546 | // |
duke@435 | 1547 | // Output: |
duke@435 | 1548 | // rax, == 0 - success |
duke@435 | 1549 | // rax, == -1 - need to call System.arraycopy |
duke@435 | 1550 | // |
duke@435 | 1551 | // Examines the alignment of the operands and dispatches |
duke@435 | 1552 | // to a long, int, short, or byte copy loop. |
duke@435 | 1553 | // |
duke@435 | 1554 | address generate_unsafe_copy(const char *name, |
duke@435 | 1555 | address byte_copy_entry, |
duke@435 | 1556 | address short_copy_entry, |
duke@435 | 1557 | address int_copy_entry, |
duke@435 | 1558 | address long_copy_entry) { |
duke@435 | 1559 | |
duke@435 | 1560 | Label L_long_aligned, L_int_aligned, L_short_aligned; |
duke@435 | 1561 | |
duke@435 | 1562 | __ align(CodeEntryAlignment); |
duke@435 | 1563 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1564 | address start = __ pc(); |
duke@435 | 1565 | |
duke@435 | 1566 | const Register from = rax; // source array address |
duke@435 | 1567 | const Register to = rdx; // destination array address |
duke@435 | 1568 | const Register count = rcx; // elements count |
duke@435 | 1569 | |
duke@435 | 1570 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
never@739 | 1571 | __ push(rsi); |
never@739 | 1572 | __ push(rdi); |
duke@435 | 1573 | Address from_arg(rsp, 12+ 4); // from |
duke@435 | 1574 | Address to_arg(rsp, 12+ 8); // to |
duke@435 | 1575 | Address count_arg(rsp, 12+12); // byte count |
duke@435 | 1576 | |
duke@435 | 1577 | // Load up: |
never@739 | 1578 | __ movptr(from , from_arg); |
never@739 | 1579 | __ movptr(to , to_arg); |
never@739 | 1580 | __ movl2ptr(count, count_arg); |
duke@435 | 1581 | |
duke@435 | 1582 | // bump this on entry, not on exit: |
duke@435 | 1583 | inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr); |
duke@435 | 1584 | |
duke@435 | 1585 | const Register bits = rsi; |
never@739 | 1586 | __ mov(bits, from); |
never@739 | 1587 | __ orptr(bits, to); |
never@739 | 1588 | __ orptr(bits, count); |
duke@435 | 1589 | |
duke@435 | 1590 | __ testl(bits, BytesPerLong-1); |
duke@435 | 1591 | __ jccb(Assembler::zero, L_long_aligned); |
duke@435 | 1592 | |
duke@435 | 1593 | __ testl(bits, BytesPerInt-1); |
duke@435 | 1594 | __ jccb(Assembler::zero, L_int_aligned); |
duke@435 | 1595 | |
duke@435 | 1596 | __ testl(bits, BytesPerShort-1); |
duke@435 | 1597 | __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry)); |
duke@435 | 1598 | |
duke@435 | 1599 | __ BIND(L_short_aligned); |
never@739 | 1600 | __ shrptr(count, LogBytesPerShort); // size => short_count |
duke@435 | 1601 | __ movl(count_arg, count); // update 'count' |
duke@435 | 1602 | __ jump(RuntimeAddress(short_copy_entry)); |
duke@435 | 1603 | |
duke@435 | 1604 | __ BIND(L_int_aligned); |
never@739 | 1605 | __ shrptr(count, LogBytesPerInt); // size => int_count |
duke@435 | 1606 | __ movl(count_arg, count); // update 'count' |
duke@435 | 1607 | __ jump(RuntimeAddress(int_copy_entry)); |
duke@435 | 1608 | |
duke@435 | 1609 | __ BIND(L_long_aligned); |
never@739 | 1610 | __ shrptr(count, LogBytesPerLong); // size => qword_count |
duke@435 | 1611 | __ movl(count_arg, count); // update 'count' |
never@739 | 1612 | __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it. |
never@739 | 1613 | __ pop(rsi); |
duke@435 | 1614 | __ jump(RuntimeAddress(long_copy_entry)); |
duke@435 | 1615 | |
duke@435 | 1616 | return start; |
duke@435 | 1617 | } |
duke@435 | 1618 | |
duke@435 | 1619 | |
duke@435 | 1620 | // Perform range checks on the proposed arraycopy. |
duke@435 | 1621 | // Smashes src_pos and dst_pos. (Uses them up for temps.) |
duke@435 | 1622 | void arraycopy_range_checks(Register src, |
duke@435 | 1623 | Register src_pos, |
duke@435 | 1624 | Register dst, |
duke@435 | 1625 | Register dst_pos, |
duke@435 | 1626 | Address& length, |
duke@435 | 1627 | Label& L_failed) { |
duke@435 | 1628 | BLOCK_COMMENT("arraycopy_range_checks:"); |
duke@435 | 1629 | const Register src_end = src_pos; // source array end position |
duke@435 | 1630 | const Register dst_end = dst_pos; // destination array end position |
duke@435 | 1631 | __ addl(src_end, length); // src_pos + length |
duke@435 | 1632 | __ addl(dst_end, length); // dst_pos + length |
duke@435 | 1633 | |
duke@435 | 1634 | // if (src_pos + length > arrayOop(src)->length() ) FAIL; |
duke@435 | 1635 | __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes())); |
duke@435 | 1636 | __ jcc(Assembler::above, L_failed); |
duke@435 | 1637 | |
duke@435 | 1638 | // if (dst_pos + length > arrayOop(dst)->length() ) FAIL; |
duke@435 | 1639 | __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes())); |
duke@435 | 1640 | __ jcc(Assembler::above, L_failed); |
duke@435 | 1641 | |
duke@435 | 1642 | BLOCK_COMMENT("arraycopy_range_checks done"); |
duke@435 | 1643 | } |
duke@435 | 1644 | |
duke@435 | 1645 | |
duke@435 | 1646 | // |
duke@435 | 1647 | // Generate generic array copy stubs |
duke@435 | 1648 | // |
duke@435 | 1649 | // Input: |
duke@435 | 1650 | // 4(rsp) - src oop |
duke@435 | 1651 | // 8(rsp) - src_pos |
duke@435 | 1652 | // 12(rsp) - dst oop |
duke@435 | 1653 | // 16(rsp) - dst_pos |
duke@435 | 1654 | // 20(rsp) - element count |
duke@435 | 1655 | // |
duke@435 | 1656 | // Output: |
duke@435 | 1657 | // rax, == 0 - success |
duke@435 | 1658 | // rax, == -1^K - failure, where K is partial transfer count |
duke@435 | 1659 | // |
duke@435 | 1660 | address generate_generic_copy(const char *name, |
duke@435 | 1661 | address entry_jbyte_arraycopy, |
duke@435 | 1662 | address entry_jshort_arraycopy, |
duke@435 | 1663 | address entry_jint_arraycopy, |
duke@435 | 1664 | address entry_oop_arraycopy, |
duke@435 | 1665 | address entry_jlong_arraycopy, |
duke@435 | 1666 | address entry_checkcast_arraycopy) { |
duke@435 | 1667 | Label L_failed, L_failed_0, L_objArray; |
duke@435 | 1668 | |
duke@435 | 1669 | { int modulus = CodeEntryAlignment; |
duke@435 | 1670 | int target = modulus - 5; // 5 = sizeof jmp(L_failed) |
duke@435 | 1671 | int advance = target - (__ offset() % modulus); |
duke@435 | 1672 | if (advance < 0) advance += modulus; |
duke@435 | 1673 | if (advance > 0) __ nop(advance); |
duke@435 | 1674 | } |
duke@435 | 1675 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1676 | |
duke@435 | 1677 | // Short-hop target to L_failed. Makes for denser prologue code. |
duke@435 | 1678 | __ BIND(L_failed_0); |
duke@435 | 1679 | __ jmp(L_failed); |
duke@435 | 1680 | assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed"); |
duke@435 | 1681 | |
duke@435 | 1682 | __ align(CodeEntryAlignment); |
duke@435 | 1683 | address start = __ pc(); |
duke@435 | 1684 | |
duke@435 | 1685 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
never@739 | 1686 | __ push(rsi); |
never@739 | 1687 | __ push(rdi); |
duke@435 | 1688 | |
duke@435 | 1689 | // bump this on entry, not on exit: |
duke@435 | 1690 | inc_counter_np(SharedRuntime::_generic_array_copy_ctr); |
duke@435 | 1691 | |
duke@435 | 1692 | // Input values |
duke@435 | 1693 | Address SRC (rsp, 12+ 4); |
duke@435 | 1694 | Address SRC_POS (rsp, 12+ 8); |
duke@435 | 1695 | Address DST (rsp, 12+12); |
duke@435 | 1696 | Address DST_POS (rsp, 12+16); |
duke@435 | 1697 | Address LENGTH (rsp, 12+20); |
duke@435 | 1698 | |
duke@435 | 1699 | //----------------------------------------------------------------------- |
duke@435 | 1700 | // Assembler stub will be used for this call to arraycopy |
duke@435 | 1701 | // if the following conditions are met: |
duke@435 | 1702 | // |
duke@435 | 1703 | // (1) src and dst must not be null. |
duke@435 | 1704 | // (2) src_pos must not be negative. |
duke@435 | 1705 | // (3) dst_pos must not be negative. |
duke@435 | 1706 | // (4) length must not be negative. |
duke@435 | 1707 | // (5) src klass and dst klass should be the same and not NULL. |
duke@435 | 1708 | // (6) src and dst should be arrays. |
duke@435 | 1709 | // (7) src_pos + length must not exceed length of src. |
duke@435 | 1710 | // (8) dst_pos + length must not exceed length of dst. |
duke@435 | 1711 | // |
duke@435 | 1712 | |
duke@435 | 1713 | const Register src = rax; // source array oop |
duke@435 | 1714 | const Register src_pos = rsi; |
duke@435 | 1715 | const Register dst = rdx; // destination array oop |
duke@435 | 1716 | const Register dst_pos = rdi; |
duke@435 | 1717 | const Register length = rcx; // transfer count |
duke@435 | 1718 | |
duke@435 | 1719 | // if (src == NULL) return -1; |
never@739 | 1720 | __ movptr(src, SRC); // src oop |
never@739 | 1721 | __ testptr(src, src); |
duke@435 | 1722 | __ jccb(Assembler::zero, L_failed_0); |
duke@435 | 1723 | |
duke@435 | 1724 | // if (src_pos < 0) return -1; |
never@739 | 1725 | __ movl2ptr(src_pos, SRC_POS); // src_pos |
duke@435 | 1726 | __ testl(src_pos, src_pos); |
duke@435 | 1727 | __ jccb(Assembler::negative, L_failed_0); |
duke@435 | 1728 | |
duke@435 | 1729 | // if (dst == NULL) return -1; |
never@739 | 1730 | __ movptr(dst, DST); // dst oop |
never@739 | 1731 | __ testptr(dst, dst); |
duke@435 | 1732 | __ jccb(Assembler::zero, L_failed_0); |
duke@435 | 1733 | |
duke@435 | 1734 | // if (dst_pos < 0) return -1; |
never@739 | 1735 | __ movl2ptr(dst_pos, DST_POS); // dst_pos |
duke@435 | 1736 | __ testl(dst_pos, dst_pos); |
duke@435 | 1737 | __ jccb(Assembler::negative, L_failed_0); |
duke@435 | 1738 | |
duke@435 | 1739 | // if (length < 0) return -1; |
never@739 | 1740 | __ movl2ptr(length, LENGTH); // length |
duke@435 | 1741 | __ testl(length, length); |
duke@435 | 1742 | __ jccb(Assembler::negative, L_failed_0); |
duke@435 | 1743 | |
duke@435 | 1744 | // if (src->klass() == NULL) return -1; |
duke@435 | 1745 | Address src_klass_addr(src, oopDesc::klass_offset_in_bytes()); |
duke@435 | 1746 | Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes()); |
duke@435 | 1747 | const Register rcx_src_klass = rcx; // array klass |
never@739 | 1748 | __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes())); |
duke@435 | 1749 | |
duke@435 | 1750 | #ifdef ASSERT |
duke@435 | 1751 | // assert(src->klass() != NULL); |
duke@435 | 1752 | BLOCK_COMMENT("assert klasses not null"); |
duke@435 | 1753 | { Label L1, L2; |
never@739 | 1754 | __ testptr(rcx_src_klass, rcx_src_klass); |
duke@435 | 1755 | __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL |
duke@435 | 1756 | __ bind(L1); |
duke@435 | 1757 | __ stop("broken null klass"); |
duke@435 | 1758 | __ bind(L2); |
never@739 | 1759 | __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD); |
duke@435 | 1760 | __ jccb(Assembler::equal, L1); // this would be broken also |
duke@435 | 1761 | BLOCK_COMMENT("assert done"); |
duke@435 | 1762 | } |
duke@435 | 1763 | #endif //ASSERT |
duke@435 | 1764 | |
duke@435 | 1765 | // Load layout helper (32-bits) |
duke@435 | 1766 | // |
duke@435 | 1767 | // |array_tag| | header_size | element_type | |log2_element_size| |
duke@435 | 1768 | // 32 30 24 16 8 2 0 |
duke@435 | 1769 | // |
duke@435 | 1770 | // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 |
duke@435 | 1771 | // |
duke@435 | 1772 | |
stefank@3391 | 1773 | int lh_offset = in_bytes(Klass::layout_helper_offset()); |
duke@435 | 1774 | Address src_klass_lh_addr(rcx_src_klass, lh_offset); |
duke@435 | 1775 | |
duke@435 | 1776 | // Handle objArrays completely differently... |
duke@435 | 1777 | jint objArray_lh = Klass::array_layout_helper(T_OBJECT); |
duke@435 | 1778 | __ cmpl(src_klass_lh_addr, objArray_lh); |
duke@435 | 1779 | __ jcc(Assembler::equal, L_objArray); |
duke@435 | 1780 | |
duke@435 | 1781 | // if (src->klass() != dst->klass()) return -1; |
never@739 | 1782 | __ cmpptr(rcx_src_klass, dst_klass_addr); |
duke@435 | 1783 | __ jccb(Assembler::notEqual, L_failed_0); |
duke@435 | 1784 | |
duke@435 | 1785 | const Register rcx_lh = rcx; // layout helper |
duke@435 | 1786 | assert(rcx_lh == rcx_src_klass, "known alias"); |
duke@435 | 1787 | __ movl(rcx_lh, src_klass_lh_addr); |
duke@435 | 1788 | |
duke@435 | 1789 | // if (!src->is_Array()) return -1; |
duke@435 | 1790 | __ cmpl(rcx_lh, Klass::_lh_neutral_value); |
duke@435 | 1791 | __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp |
duke@435 | 1792 | |
duke@435 | 1793 | // At this point, it is known to be a typeArray (array_tag 0x3). |
duke@435 | 1794 | #ifdef ASSERT |
duke@435 | 1795 | { Label L; |
duke@435 | 1796 | __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift)); |
duke@435 | 1797 | __ jcc(Assembler::greaterEqual, L); // signed cmp |
duke@435 | 1798 | __ stop("must be a primitive array"); |
duke@435 | 1799 | __ bind(L); |
duke@435 | 1800 | } |
duke@435 | 1801 | #endif |
duke@435 | 1802 | |
duke@435 | 1803 | assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh); |
duke@435 | 1804 | arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); |
duke@435 | 1805 | |
coleenp@4142 | 1806 | // TypeArrayKlass |
duke@435 | 1807 | // |
duke@435 | 1808 | // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); |
duke@435 | 1809 | // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); |
duke@435 | 1810 | // |
duke@435 | 1811 | const Register rsi_offset = rsi; // array offset |
duke@435 | 1812 | const Register src_array = src; // src array offset |
duke@435 | 1813 | const Register dst_array = dst; // dst array offset |
duke@435 | 1814 | const Register rdi_elsize = rdi; // log2 element size |
duke@435 | 1815 | |
never@739 | 1816 | __ mov(rsi_offset, rcx_lh); |
never@739 | 1817 | __ shrptr(rsi_offset, Klass::_lh_header_size_shift); |
never@739 | 1818 | __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset |
never@739 | 1819 | __ addptr(src_array, rsi_offset); // src array offset |
never@739 | 1820 | __ addptr(dst_array, rsi_offset); // dst array offset |
never@739 | 1821 | __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize |
duke@435 | 1822 | |
duke@435 | 1823 | // next registers should be set before the jump to corresponding stub |
duke@435 | 1824 | const Register from = src; // source array address |
duke@435 | 1825 | const Register to = dst; // destination array address |
duke@435 | 1826 | const Register count = rcx; // elements count |
duke@435 | 1827 | // some of them should be duplicated on stack |
duke@435 | 1828 | #define FROM Address(rsp, 12+ 4) |
duke@435 | 1829 | #define TO Address(rsp, 12+ 8) // Not used now |
duke@435 | 1830 | #define COUNT Address(rsp, 12+12) // Only for oop arraycopy |
duke@435 | 1831 | |
duke@435 | 1832 | BLOCK_COMMENT("scale indexes to element size"); |
never@739 | 1833 | __ movl2ptr(rsi, SRC_POS); // src_pos |
never@739 | 1834 | __ shlptr(rsi); // src_pos << rcx (log2 elsize) |
duke@435 | 1835 | assert(src_array == from, ""); |
never@739 | 1836 | __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize |
never@739 | 1837 | __ movl2ptr(rdi, DST_POS); // dst_pos |
never@739 | 1838 | __ shlptr(rdi); // dst_pos << rcx (log2 elsize) |
duke@435 | 1839 | assert(dst_array == to, ""); |
never@739 | 1840 | __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize |
never@739 | 1841 | __ movptr(FROM, from); // src_addr |
never@739 | 1842 | __ mov(rdi_elsize, rcx_lh); // log2 elsize |
never@739 | 1843 | __ movl2ptr(count, LENGTH); // elements count |
duke@435 | 1844 | |
duke@435 | 1845 | BLOCK_COMMENT("choose copy loop based on element size"); |
duke@435 | 1846 | __ cmpl(rdi_elsize, 0); |
duke@435 | 1847 | |
duke@435 | 1848 | __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy)); |
duke@435 | 1849 | __ cmpl(rdi_elsize, LogBytesPerShort); |
duke@435 | 1850 | __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy)); |
duke@435 | 1851 | __ cmpl(rdi_elsize, LogBytesPerInt); |
duke@435 | 1852 | __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy)); |
duke@435 | 1853 | #ifdef ASSERT |
duke@435 | 1854 | __ cmpl(rdi_elsize, LogBytesPerLong); |
duke@435 | 1855 | __ jccb(Assembler::notEqual, L_failed); |
duke@435 | 1856 | #endif |
never@739 | 1857 | __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it. |
never@739 | 1858 | __ pop(rsi); |
duke@435 | 1859 | __ jump(RuntimeAddress(entry_jlong_arraycopy)); |
duke@435 | 1860 | |
duke@435 | 1861 | __ BIND(L_failed); |
never@739 | 1862 | __ xorptr(rax, rax); |
never@739 | 1863 | __ notptr(rax); // return -1 |
never@739 | 1864 | __ pop(rdi); |
never@739 | 1865 | __ pop(rsi); |
duke@435 | 1866 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 1867 | __ ret(0); |
duke@435 | 1868 | |
coleenp@4142 | 1869 | // ObjArrayKlass |
duke@435 | 1870 | __ BIND(L_objArray); |
duke@435 | 1871 | // live at this point: rcx_src_klass, src[_pos], dst[_pos] |
duke@435 | 1872 | |
duke@435 | 1873 | Label L_plain_copy, L_checkcast_copy; |
duke@435 | 1874 | // test array classes for subtyping |
never@739 | 1875 | __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality |
duke@435 | 1876 | __ jccb(Assembler::notEqual, L_checkcast_copy); |
duke@435 | 1877 | |
duke@435 | 1878 | // Identically typed arrays can be copied without element-wise checks. |
duke@435 | 1879 | assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass); |
duke@435 | 1880 | arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); |
duke@435 | 1881 | |
duke@435 | 1882 | __ BIND(L_plain_copy); |
never@739 | 1883 | __ movl2ptr(count, LENGTH); // elements count |
never@739 | 1884 | __ movl2ptr(src_pos, SRC_POS); // reload src_pos |
never@739 | 1885 | __ lea(from, Address(src, src_pos, Address::times_ptr, |
never@739 | 1886 | arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr |
never@739 | 1887 | __ movl2ptr(dst_pos, DST_POS); // reload dst_pos |
never@739 | 1888 | __ lea(to, Address(dst, dst_pos, Address::times_ptr, |
never@739 | 1889 | arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr |
never@739 | 1890 | __ movptr(FROM, from); // src_addr |
never@739 | 1891 | __ movptr(TO, to); // dst_addr |
duke@435 | 1892 | __ movl(COUNT, count); // count |
duke@435 | 1893 | __ jump(RuntimeAddress(entry_oop_arraycopy)); |
duke@435 | 1894 | |
duke@435 | 1895 | __ BIND(L_checkcast_copy); |
duke@435 | 1896 | // live at this point: rcx_src_klass, dst[_pos], src[_pos] |
duke@435 | 1897 | { |
duke@435 | 1898 | // Handy offsets: |
coleenp@4142 | 1899 | int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset()); |
stefank@3391 | 1900 | int sco_offset = in_bytes(Klass::super_check_offset_offset()); |
duke@435 | 1901 | |
duke@435 | 1902 | Register rsi_dst_klass = rsi; |
duke@435 | 1903 | Register rdi_temp = rdi; |
duke@435 | 1904 | assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos"); |
duke@435 | 1905 | assert(rdi_temp == dst_pos, "expected alias w/ dst_pos"); |
duke@435 | 1906 | Address dst_klass_lh_addr(rsi_dst_klass, lh_offset); |
duke@435 | 1907 | |
duke@435 | 1908 | // Before looking at dst.length, make sure dst is also an objArray. |
never@739 | 1909 | __ movptr(rsi_dst_klass, dst_klass_addr); |
duke@435 | 1910 | __ cmpl(dst_klass_lh_addr, objArray_lh); |
duke@435 | 1911 | __ jccb(Assembler::notEqual, L_failed); |
duke@435 | 1912 | |
duke@435 | 1913 | // It is safe to examine both src.length and dst.length. |
never@739 | 1914 | __ movl2ptr(src_pos, SRC_POS); // reload rsi |
duke@435 | 1915 | arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); |
duke@435 | 1916 | // (Now src_pos and dst_pos are killed, but not src and dst.) |
duke@435 | 1917 | |
duke@435 | 1918 | // We'll need this temp (don't forget to pop it after the type check). |
never@739 | 1919 | __ push(rbx); |
duke@435 | 1920 | Register rbx_src_klass = rbx; |
duke@435 | 1921 | |
never@739 | 1922 | __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx |
never@739 | 1923 | __ movptr(rsi_dst_klass, dst_klass_addr); |
duke@435 | 1924 | Address super_check_offset_addr(rsi_dst_klass, sco_offset); |
duke@435 | 1925 | Label L_fail_array_check; |
duke@435 | 1926 | generate_type_check(rbx_src_klass, |
duke@435 | 1927 | super_check_offset_addr, dst_klass_addr, |
duke@435 | 1928 | rdi_temp, NULL, &L_fail_array_check); |
duke@435 | 1929 | // (On fall-through, we have passed the array type check.) |
never@739 | 1930 | __ pop(rbx); |
duke@435 | 1931 | __ jmp(L_plain_copy); |
duke@435 | 1932 | |
duke@435 | 1933 | __ BIND(L_fail_array_check); |
duke@435 | 1934 | // Reshuffle arguments so we can call checkcast_arraycopy: |
duke@435 | 1935 | |
duke@435 | 1936 | // match initial saves for checkcast_arraycopy |
never@739 | 1937 | // push(rsi); // already done; see above |
never@739 | 1938 | // push(rdi); // already done; see above |
never@739 | 1939 | // push(rbx); // already done; see above |
duke@435 | 1940 | |
duke@435 | 1941 | // Marshal outgoing arguments now, freeing registers. |
duke@435 | 1942 | Address from_arg(rsp, 16+ 4); // from |
duke@435 | 1943 | Address to_arg(rsp, 16+ 8); // to |
duke@435 | 1944 | Address length_arg(rsp, 16+12); // elements count |
duke@435 | 1945 | Address ckoff_arg(rsp, 16+16); // super_check_offset |
duke@435 | 1946 | Address ckval_arg(rsp, 16+20); // super_klass |
duke@435 | 1947 | |
duke@435 | 1948 | Address SRC_POS_arg(rsp, 16+ 8); |
duke@435 | 1949 | Address DST_POS_arg(rsp, 16+16); |
duke@435 | 1950 | Address LENGTH_arg(rsp, 16+20); |
duke@435 | 1951 | // push rbx, changed the incoming offsets (why not just use rbp,??) |
duke@435 | 1952 | // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, ""); |
duke@435 | 1953 | |
never@739 | 1954 | __ movptr(rbx, Address(rsi_dst_klass, ek_offset)); |
never@739 | 1955 | __ movl2ptr(length, LENGTH_arg); // reload elements count |
never@739 | 1956 | __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos |
never@739 | 1957 | __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos |
duke@435 | 1958 | |
never@739 | 1959 | __ movptr(ckval_arg, rbx); // destination element type |
duke@435 | 1960 | __ movl(rbx, Address(rbx, sco_offset)); |
duke@435 | 1961 | __ movl(ckoff_arg, rbx); // corresponding class check offset |
duke@435 | 1962 | |
duke@435 | 1963 | __ movl(length_arg, length); // outgoing length argument |
duke@435 | 1964 | |
never@739 | 1965 | __ lea(from, Address(src, src_pos, Address::times_ptr, |
duke@435 | 1966 | arrayOopDesc::base_offset_in_bytes(T_OBJECT))); |
never@739 | 1967 | __ movptr(from_arg, from); |
duke@435 | 1968 | |
never@739 | 1969 | __ lea(to, Address(dst, dst_pos, Address::times_ptr, |
duke@435 | 1970 | arrayOopDesc::base_offset_in_bytes(T_OBJECT))); |
never@739 | 1971 | __ movptr(to_arg, to); |
duke@435 | 1972 | __ jump(RuntimeAddress(entry_checkcast_arraycopy)); |
duke@435 | 1973 | } |
duke@435 | 1974 | |
duke@435 | 1975 | return start; |
duke@435 | 1976 | } |
duke@435 | 1977 | |
duke@435 | 1978 | void generate_arraycopy_stubs() { |
duke@435 | 1979 | address entry; |
duke@435 | 1980 | address entry_jbyte_arraycopy; |
duke@435 | 1981 | address entry_jshort_arraycopy; |
duke@435 | 1982 | address entry_jint_arraycopy; |
duke@435 | 1983 | address entry_oop_arraycopy; |
duke@435 | 1984 | address entry_jlong_arraycopy; |
duke@435 | 1985 | address entry_checkcast_arraycopy; |
duke@435 | 1986 | |
duke@435 | 1987 | StubRoutines::_arrayof_jbyte_disjoint_arraycopy = |
duke@435 | 1988 | generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry, |
duke@435 | 1989 | "arrayof_jbyte_disjoint_arraycopy"); |
duke@435 | 1990 | StubRoutines::_arrayof_jbyte_arraycopy = |
duke@435 | 1991 | generate_conjoint_copy(T_BYTE, true, Address::times_1, entry, |
duke@435 | 1992 | NULL, "arrayof_jbyte_arraycopy"); |
duke@435 | 1993 | StubRoutines::_jbyte_disjoint_arraycopy = |
duke@435 | 1994 | generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry, |
duke@435 | 1995 | "jbyte_disjoint_arraycopy"); |
duke@435 | 1996 | StubRoutines::_jbyte_arraycopy = |
duke@435 | 1997 | generate_conjoint_copy(T_BYTE, false, Address::times_1, entry, |
duke@435 | 1998 | &entry_jbyte_arraycopy, "jbyte_arraycopy"); |
duke@435 | 1999 | |
duke@435 | 2000 | StubRoutines::_arrayof_jshort_disjoint_arraycopy = |
duke@435 | 2001 | generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry, |
duke@435 | 2002 | "arrayof_jshort_disjoint_arraycopy"); |
duke@435 | 2003 | StubRoutines::_arrayof_jshort_arraycopy = |
duke@435 | 2004 | generate_conjoint_copy(T_SHORT, true, Address::times_2, entry, |
duke@435 | 2005 | NULL, "arrayof_jshort_arraycopy"); |
duke@435 | 2006 | StubRoutines::_jshort_disjoint_arraycopy = |
duke@435 | 2007 | generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry, |
duke@435 | 2008 | "jshort_disjoint_arraycopy"); |
duke@435 | 2009 | StubRoutines::_jshort_arraycopy = |
duke@435 | 2010 | generate_conjoint_copy(T_SHORT, false, Address::times_2, entry, |
duke@435 | 2011 | &entry_jshort_arraycopy, "jshort_arraycopy"); |
duke@435 | 2012 | |
duke@435 | 2013 | // Next arrays are always aligned on 4 bytes at least. |
duke@435 | 2014 | StubRoutines::_jint_disjoint_arraycopy = |
duke@435 | 2015 | generate_disjoint_copy(T_INT, true, Address::times_4, &entry, |
duke@435 | 2016 | "jint_disjoint_arraycopy"); |
duke@435 | 2017 | StubRoutines::_jint_arraycopy = |
duke@435 | 2018 | generate_conjoint_copy(T_INT, true, Address::times_4, entry, |
duke@435 | 2019 | &entry_jint_arraycopy, "jint_arraycopy"); |
duke@435 | 2020 | |
duke@435 | 2021 | StubRoutines::_oop_disjoint_arraycopy = |
never@739 | 2022 | generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry, |
duke@435 | 2023 | "oop_disjoint_arraycopy"); |
duke@435 | 2024 | StubRoutines::_oop_arraycopy = |
never@739 | 2025 | generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry, |
duke@435 | 2026 | &entry_oop_arraycopy, "oop_arraycopy"); |
duke@435 | 2027 | |
iveresov@2606 | 2028 | StubRoutines::_oop_disjoint_arraycopy_uninit = |
iveresov@2606 | 2029 | generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry, |
iveresov@2606 | 2030 | "oop_disjoint_arraycopy_uninit", |
iveresov@2606 | 2031 | /*dest_uninitialized*/true); |
iveresov@2606 | 2032 | StubRoutines::_oop_arraycopy_uninit = |
iveresov@2606 | 2033 | generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry, |
iveresov@2606 | 2034 | NULL, "oop_arraycopy_uninit", |
iveresov@2606 | 2035 | /*dest_uninitialized*/true); |
iveresov@2606 | 2036 | |
duke@435 | 2037 | StubRoutines::_jlong_disjoint_arraycopy = |
duke@435 | 2038 | generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy"); |
duke@435 | 2039 | StubRoutines::_jlong_arraycopy = |
duke@435 | 2040 | generate_conjoint_long_copy(entry, &entry_jlong_arraycopy, |
duke@435 | 2041 | "jlong_arraycopy"); |
duke@435 | 2042 | |
never@2118 | 2043 | StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill"); |
never@2118 | 2044 | StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill"); |
never@2118 | 2045 | StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill"); |
never@2118 | 2046 | StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill"); |
never@2118 | 2047 | StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill"); |
never@2118 | 2048 | StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill"); |
never@2118 | 2049 | |
iveresov@2606 | 2050 | StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy; |
iveresov@2606 | 2051 | StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy; |
iveresov@2606 | 2052 | StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit; |
iveresov@2606 | 2053 | StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy; |
duke@435 | 2054 | |
iveresov@2606 | 2055 | StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy; |
iveresov@2606 | 2056 | StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; |
iveresov@2606 | 2057 | StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit; |
iveresov@2606 | 2058 | StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy; |
duke@435 | 2059 | |
duke@435 | 2060 | StubRoutines::_checkcast_arraycopy = |
iveresov@2606 | 2061 | generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy); |
iveresov@2606 | 2062 | StubRoutines::_checkcast_arraycopy_uninit = |
iveresov@2606 | 2063 | generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true); |
duke@435 | 2064 | |
duke@435 | 2065 | StubRoutines::_unsafe_arraycopy = |
duke@435 | 2066 | generate_unsafe_copy("unsafe_arraycopy", |
duke@435 | 2067 | entry_jbyte_arraycopy, |
duke@435 | 2068 | entry_jshort_arraycopy, |
duke@435 | 2069 | entry_jint_arraycopy, |
duke@435 | 2070 | entry_jlong_arraycopy); |
duke@435 | 2071 | |
duke@435 | 2072 | StubRoutines::_generic_arraycopy = |
duke@435 | 2073 | generate_generic_copy("generic_arraycopy", |
duke@435 | 2074 | entry_jbyte_arraycopy, |
duke@435 | 2075 | entry_jshort_arraycopy, |
duke@435 | 2076 | entry_jint_arraycopy, |
duke@435 | 2077 | entry_oop_arraycopy, |
duke@435 | 2078 | entry_jlong_arraycopy, |
duke@435 | 2079 | entry_checkcast_arraycopy); |
duke@435 | 2080 | } |
duke@435 | 2081 | |
never@1609 | 2082 | void generate_math_stubs() { |
never@1609 | 2083 | { |
never@1609 | 2084 | StubCodeMark mark(this, "StubRoutines", "log"); |
never@1609 | 2085 | StubRoutines::_intrinsic_log = (double (*)(double)) __ pc(); |
never@1609 | 2086 | |
never@1609 | 2087 | __ fld_d(Address(rsp, 4)); |
never@1609 | 2088 | __ flog(); |
never@1609 | 2089 | __ ret(0); |
never@1609 | 2090 | } |
never@1609 | 2091 | { |
never@1609 | 2092 | StubCodeMark mark(this, "StubRoutines", "log10"); |
never@1609 | 2093 | StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc(); |
never@1609 | 2094 | |
never@1609 | 2095 | __ fld_d(Address(rsp, 4)); |
never@1609 | 2096 | __ flog10(); |
never@1609 | 2097 | __ ret(0); |
never@1609 | 2098 | } |
never@1609 | 2099 | { |
never@1609 | 2100 | StubCodeMark mark(this, "StubRoutines", "sin"); |
never@1609 | 2101 | StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc(); |
never@1609 | 2102 | |
never@1609 | 2103 | __ fld_d(Address(rsp, 4)); |
never@1609 | 2104 | __ trigfunc('s'); |
never@1609 | 2105 | __ ret(0); |
never@1609 | 2106 | } |
never@1609 | 2107 | { |
never@1609 | 2108 | StubCodeMark mark(this, "StubRoutines", "cos"); |
never@1609 | 2109 | StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc(); |
never@1609 | 2110 | |
never@1609 | 2111 | __ fld_d(Address(rsp, 4)); |
never@1609 | 2112 | __ trigfunc('c'); |
never@1609 | 2113 | __ ret(0); |
never@1609 | 2114 | } |
never@1609 | 2115 | { |
never@1609 | 2116 | StubCodeMark mark(this, "StubRoutines", "tan"); |
never@1609 | 2117 | StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc(); |
never@1609 | 2118 | |
never@1609 | 2119 | __ fld_d(Address(rsp, 4)); |
never@1609 | 2120 | __ trigfunc('t'); |
never@1609 | 2121 | __ ret(0); |
never@1609 | 2122 | } |
roland@3787 | 2123 | { |
roland@3787 | 2124 | StubCodeMark mark(this, "StubRoutines", "exp"); |
roland@3787 | 2125 | StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc(); |
never@1609 | 2126 | |
roland@3787 | 2127 | __ fld_d(Address(rsp, 4)); |
roland@3787 | 2128 | __ exp_with_fallback(0); |
roland@3787 | 2129 | __ ret(0); |
roland@3787 | 2130 | } |
roland@3787 | 2131 | { |
roland@3787 | 2132 | StubCodeMark mark(this, "StubRoutines", "pow"); |
roland@3787 | 2133 | StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc(); |
roland@3787 | 2134 | |
roland@3787 | 2135 | __ fld_d(Address(rsp, 12)); |
roland@3787 | 2136 | __ fld_d(Address(rsp, 4)); |
roland@3787 | 2137 | __ pow_with_fallback(0); |
roland@3787 | 2138 | __ ret(0); |
roland@3787 | 2139 | } |
never@1609 | 2140 | } |
never@1609 | 2141 | |
kvn@4205 | 2142 | // AES intrinsic stubs |
kvn@4205 | 2143 | enum {AESBlockSize = 16}; |
kvn@4205 | 2144 | |
kvn@4205 | 2145 | address generate_key_shuffle_mask() { |
kvn@4205 | 2146 | __ align(16); |
kvn@4205 | 2147 | StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask"); |
kvn@4205 | 2148 | address start = __ pc(); |
kvn@4205 | 2149 | __ emit_data(0x00010203, relocInfo::none, 0 ); |
kvn@4205 | 2150 | __ emit_data(0x04050607, relocInfo::none, 0 ); |
kvn@4205 | 2151 | __ emit_data(0x08090a0b, relocInfo::none, 0 ); |
kvn@4205 | 2152 | __ emit_data(0x0c0d0e0f, relocInfo::none, 0 ); |
kvn@4205 | 2153 | return start; |
kvn@4205 | 2154 | } |
kvn@4205 | 2155 | |
kvn@4205 | 2156 | // Utility routine for loading a 128-bit key word in little endian format |
kvn@4205 | 2157 | // can optionally specify that the shuffle mask is already in an xmmregister |
kvn@4205 | 2158 | void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) { |
kvn@4205 | 2159 | __ movdqu(xmmdst, Address(key, offset)); |
kvn@4205 | 2160 | if (xmm_shuf_mask != NULL) { |
kvn@4205 | 2161 | __ pshufb(xmmdst, xmm_shuf_mask); |
kvn@4205 | 2162 | } else { |
kvn@4205 | 2163 | __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); |
kvn@4205 | 2164 | } |
kvn@4205 | 2165 | } |
kvn@4205 | 2166 | |
kvn@4205 | 2167 | // aesenc using specified key+offset |
kvn@4205 | 2168 | // can optionally specify that the shuffle mask is already in an xmmregister |
kvn@4205 | 2169 | void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) { |
kvn@4205 | 2170 | load_key(xmmtmp, key, offset, xmm_shuf_mask); |
kvn@4205 | 2171 | __ aesenc(xmmdst, xmmtmp); |
kvn@4205 | 2172 | } |
kvn@4205 | 2173 | |
kvn@4205 | 2174 | // aesdec using specified key+offset |
kvn@4205 | 2175 | // can optionally specify that the shuffle mask is already in an xmmregister |
kvn@4205 | 2176 | void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) { |
kvn@4205 | 2177 | load_key(xmmtmp, key, offset, xmm_shuf_mask); |
kvn@4205 | 2178 | __ aesdec(xmmdst, xmmtmp); |
kvn@4205 | 2179 | } |
kvn@4205 | 2180 | |
kvn@4205 | 2181 | |
kvn@4205 | 2182 | // Arguments: |
kvn@4205 | 2183 | // |
kvn@4205 | 2184 | // Inputs: |
kvn@4205 | 2185 | // c_rarg0 - source byte array address |
kvn@4205 | 2186 | // c_rarg1 - destination byte array address |
kvn@4205 | 2187 | // c_rarg2 - K (key) in little endian int array |
kvn@4205 | 2188 | // |
kvn@4205 | 2189 | address generate_aescrypt_encryptBlock() { |
kvn@4363 | 2190 | assert(UseAES, "need AES instructions and misaligned SSE support"); |
kvn@4205 | 2191 | __ align(CodeEntryAlignment); |
kvn@4205 | 2192 | StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock"); |
kvn@4205 | 2193 | Label L_doLast; |
kvn@4205 | 2194 | address start = __ pc(); |
kvn@4205 | 2195 | |
kvn@4363 | 2196 | const Register from = rdx; // source array address |
kvn@4205 | 2197 | const Register to = rdx; // destination array address |
kvn@4205 | 2198 | const Register key = rcx; // key array address |
kvn@4205 | 2199 | const Register keylen = rax; |
kvn@4205 | 2200 | const Address from_param(rbp, 8+0); |
kvn@4205 | 2201 | const Address to_param (rbp, 8+4); |
kvn@4205 | 2202 | const Address key_param (rbp, 8+8); |
kvn@4205 | 2203 | |
kvn@4205 | 2204 | const XMMRegister xmm_result = xmm0; |
kvn@4363 | 2205 | const XMMRegister xmm_key_shuf_mask = xmm1; |
kvn@4363 | 2206 | const XMMRegister xmm_temp1 = xmm2; |
kvn@4363 | 2207 | const XMMRegister xmm_temp2 = xmm3; |
kvn@4363 | 2208 | const XMMRegister xmm_temp3 = xmm4; |
kvn@4363 | 2209 | const XMMRegister xmm_temp4 = xmm5; |
kvn@4363 | 2210 | |
kvn@4363 | 2211 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
kvn@4363 | 2212 | __ movptr(from, from_param); |
kvn@4363 | 2213 | __ movptr(key, key_param); |
kvn@4363 | 2214 | |
kvn@4363 | 2215 | // keylen could be only {11, 13, 15} * 4 = {44, 52, 60} |
kvn@4205 | 2216 | __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); |
kvn@4205 | 2217 | |
kvn@4205 | 2218 | __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); |
kvn@4205 | 2219 | __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input |
kvn@4363 | 2220 | __ movptr(to, to_param); |
kvn@4205 | 2221 | |
kvn@4205 | 2222 | // For encryption, the java expanded key ordering is just what we need |
kvn@4205 | 2223 | |
kvn@4363 | 2224 | load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask); |
kvn@4363 | 2225 | __ pxor(xmm_result, xmm_temp1); |
kvn@4363 | 2226 | |
kvn@4363 | 2227 | load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask); |
kvn@4363 | 2228 | load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask); |
kvn@4363 | 2229 | load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask); |
kvn@4363 | 2230 | load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask); |
kvn@4363 | 2231 | |
kvn@4363 | 2232 | __ aesenc(xmm_result, xmm_temp1); |
kvn@4363 | 2233 | __ aesenc(xmm_result, xmm_temp2); |
kvn@4363 | 2234 | __ aesenc(xmm_result, xmm_temp3); |
kvn@4363 | 2235 | __ aesenc(xmm_result, xmm_temp4); |
kvn@4363 | 2236 | |
kvn@4363 | 2237 | load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask); |
kvn@4363 | 2238 | load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask); |
kvn@4363 | 2239 | load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask); |
kvn@4363 | 2240 | load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask); |
kvn@4363 | 2241 | |
kvn@4363 | 2242 | __ aesenc(xmm_result, xmm_temp1); |
kvn@4363 | 2243 | __ aesenc(xmm_result, xmm_temp2); |
kvn@4363 | 2244 | __ aesenc(xmm_result, xmm_temp3); |
kvn@4363 | 2245 | __ aesenc(xmm_result, xmm_temp4); |
kvn@4363 | 2246 | |
kvn@4363 | 2247 | load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask); |
kvn@4363 | 2248 | load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask); |
kvn@4363 | 2249 | |
kvn@4363 | 2250 | __ cmpl(keylen, 44); |
kvn@4363 | 2251 | __ jccb(Assembler::equal, L_doLast); |
kvn@4363 | 2252 | |
kvn@4363 | 2253 | __ aesenc(xmm_result, xmm_temp1); |
kvn@4363 | 2254 | __ aesenc(xmm_result, xmm_temp2); |
kvn@4363 | 2255 | |
kvn@4363 | 2256 | load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask); |
kvn@4363 | 2257 | load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask); |
kvn@4363 | 2258 | |
kvn@4363 | 2259 | __ cmpl(keylen, 52); |
kvn@4363 | 2260 | __ jccb(Assembler::equal, L_doLast); |
kvn@4363 | 2261 | |
kvn@4363 | 2262 | __ aesenc(xmm_result, xmm_temp1); |
kvn@4363 | 2263 | __ aesenc(xmm_result, xmm_temp2); |
kvn@4363 | 2264 | |
kvn@4363 | 2265 | load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask); |
kvn@4363 | 2266 | load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask); |
kvn@4205 | 2267 | |
kvn@4205 | 2268 | __ BIND(L_doLast); |
kvn@4363 | 2269 | __ aesenc(xmm_result, xmm_temp1); |
kvn@4363 | 2270 | __ aesenclast(xmm_result, xmm_temp2); |
kvn@4205 | 2271 | __ movdqu(Address(to, 0), xmm_result); // store the result |
kvn@4205 | 2272 | __ xorptr(rax, rax); // return 0 |
kvn@4205 | 2273 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 2274 | __ ret(0); |
kvn@4205 | 2275 | |
kvn@4205 | 2276 | return start; |
kvn@4205 | 2277 | } |
kvn@4205 | 2278 | |
kvn@4205 | 2279 | |
kvn@4205 | 2280 | // Arguments: |
kvn@4205 | 2281 | // |
kvn@4205 | 2282 | // Inputs: |
kvn@4205 | 2283 | // c_rarg0 - source byte array address |
kvn@4205 | 2284 | // c_rarg1 - destination byte array address |
kvn@4205 | 2285 | // c_rarg2 - K (key) in little endian int array |
kvn@4205 | 2286 | // |
kvn@4205 | 2287 | address generate_aescrypt_decryptBlock() { |
kvn@4363 | 2288 | assert(UseAES, "need AES instructions and misaligned SSE support"); |
kvn@4205 | 2289 | __ align(CodeEntryAlignment); |
kvn@4205 | 2290 | StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock"); |
kvn@4205 | 2291 | Label L_doLast; |
kvn@4205 | 2292 | address start = __ pc(); |
kvn@4205 | 2293 | |
kvn@4363 | 2294 | const Register from = rdx; // source array address |
kvn@4205 | 2295 | const Register to = rdx; // destination array address |
kvn@4205 | 2296 | const Register key = rcx; // key array address |
kvn@4205 | 2297 | const Register keylen = rax; |
kvn@4205 | 2298 | const Address from_param(rbp, 8+0); |
kvn@4205 | 2299 | const Address to_param (rbp, 8+4); |
kvn@4205 | 2300 | const Address key_param (rbp, 8+8); |
kvn@4205 | 2301 | |
kvn@4205 | 2302 | const XMMRegister xmm_result = xmm0; |
kvn@4363 | 2303 | const XMMRegister xmm_key_shuf_mask = xmm1; |
kvn@4363 | 2304 | const XMMRegister xmm_temp1 = xmm2; |
kvn@4363 | 2305 | const XMMRegister xmm_temp2 = xmm3; |
kvn@4363 | 2306 | const XMMRegister xmm_temp3 = xmm4; |
kvn@4363 | 2307 | const XMMRegister xmm_temp4 = xmm5; |
kvn@4205 | 2308 | |
kvn@4205 | 2309 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
kvn@4363 | 2310 | __ movptr(from, from_param); |
kvn@4363 | 2311 | __ movptr(key, key_param); |
kvn@4363 | 2312 | |
kvn@4363 | 2313 | // keylen could be only {11, 13, 15} * 4 = {44, 52, 60} |
kvn@4205 | 2314 | __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); |
kvn@4205 | 2315 | |
kvn@4205 | 2316 | __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); |
kvn@4205 | 2317 | __ movdqu(xmm_result, Address(from, 0)); |
kvn@4363 | 2318 | __ movptr(to, to_param); |
kvn@4205 | 2319 | |
kvn@4205 | 2320 | // for decryption java expanded key ordering is rotated one position from what we want |
kvn@4205 | 2321 | // so we start from 0x10 here and hit 0x00 last |
kvn@4205 | 2322 | // we don't know if the key is aligned, hence not using load-execute form |
kvn@4363 | 2323 | load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask); |
kvn@4363 | 2324 | load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask); |
kvn@4363 | 2325 | load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask); |
kvn@4363 | 2326 | load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask); |
kvn@4363 | 2327 | |
kvn@4363 | 2328 | __ pxor (xmm_result, xmm_temp1); |
kvn@4363 | 2329 | __ aesdec(xmm_result, xmm_temp2); |
kvn@4363 | 2330 | __ aesdec(xmm_result, xmm_temp3); |
kvn@4363 | 2331 | __ aesdec(xmm_result, xmm_temp4); |
kvn@4363 | 2332 | |
kvn@4363 | 2333 | load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask); |
kvn@4363 | 2334 | load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask); |
kvn@4363 | 2335 | load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask); |
kvn@4363 | 2336 | load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask); |
kvn@4363 | 2337 | |
kvn@4363 | 2338 | __ aesdec(xmm_result, xmm_temp1); |
kvn@4363 | 2339 | __ aesdec(xmm_result, xmm_temp2); |
kvn@4363 | 2340 | __ aesdec(xmm_result, xmm_temp3); |
kvn@4363 | 2341 | __ aesdec(xmm_result, xmm_temp4); |
kvn@4363 | 2342 | |
kvn@4363 | 2343 | load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask); |
kvn@4363 | 2344 | load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask); |
kvn@4363 | 2345 | load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask); |
kvn@4363 | 2346 | |
kvn@4363 | 2347 | __ cmpl(keylen, 44); |
kvn@4363 | 2348 | __ jccb(Assembler::equal, L_doLast); |
kvn@4363 | 2349 | |
kvn@4363 | 2350 | __ aesdec(xmm_result, xmm_temp1); |
kvn@4363 | 2351 | __ aesdec(xmm_result, xmm_temp2); |
kvn@4363 | 2352 | |
kvn@4363 | 2353 | load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask); |
kvn@4363 | 2354 | load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask); |
kvn@4363 | 2355 | |
kvn@4363 | 2356 | __ cmpl(keylen, 52); |
kvn@4363 | 2357 | __ jccb(Assembler::equal, L_doLast); |
kvn@4363 | 2358 | |
kvn@4363 | 2359 | __ aesdec(xmm_result, xmm_temp1); |
kvn@4363 | 2360 | __ aesdec(xmm_result, xmm_temp2); |
kvn@4363 | 2361 | |
kvn@4363 | 2362 | load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask); |
kvn@4363 | 2363 | load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask); |
kvn@4205 | 2364 | |
kvn@4205 | 2365 | __ BIND(L_doLast); |
kvn@4363 | 2366 | __ aesdec(xmm_result, xmm_temp1); |
kvn@4363 | 2367 | __ aesdec(xmm_result, xmm_temp2); |
kvn@4363 | 2368 | |
kvn@4205 | 2369 | // for decryption the aesdeclast operation is always on key+0x00 |
kvn@4363 | 2370 | __ aesdeclast(xmm_result, xmm_temp3); |
kvn@4205 | 2371 | __ movdqu(Address(to, 0), xmm_result); // store the result |
kvn@4205 | 2372 | __ xorptr(rax, rax); // return 0 |
kvn@4205 | 2373 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 2374 | __ ret(0); |
kvn@4205 | 2375 | |
kvn@4205 | 2376 | return start; |
kvn@4205 | 2377 | } |
kvn@4205 | 2378 | |
kvn@4205 | 2379 | void handleSOERegisters(bool saving) { |
kvn@4205 | 2380 | const int saveFrameSizeInBytes = 4 * wordSize; |
kvn@4205 | 2381 | const Address saved_rbx (rbp, -3 * wordSize); |
kvn@4205 | 2382 | const Address saved_rsi (rbp, -2 * wordSize); |
kvn@4205 | 2383 | const Address saved_rdi (rbp, -1 * wordSize); |
kvn@4205 | 2384 | |
kvn@4205 | 2385 | if (saving) { |
kvn@4205 | 2386 | __ subptr(rsp, saveFrameSizeInBytes); |
kvn@4205 | 2387 | __ movptr(saved_rsi, rsi); |
kvn@4205 | 2388 | __ movptr(saved_rdi, rdi); |
kvn@4205 | 2389 | __ movptr(saved_rbx, rbx); |
kvn@4205 | 2390 | } else { |
kvn@4205 | 2391 | // restoring |
kvn@4205 | 2392 | __ movptr(rsi, saved_rsi); |
kvn@4205 | 2393 | __ movptr(rdi, saved_rdi); |
kvn@4205 | 2394 | __ movptr(rbx, saved_rbx); |
kvn@4205 | 2395 | } |
kvn@4205 | 2396 | } |
kvn@4205 | 2397 | |
kvn@4205 | 2398 | // Arguments: |
kvn@4205 | 2399 | // |
kvn@4205 | 2400 | // Inputs: |
kvn@4205 | 2401 | // c_rarg0 - source byte array address |
kvn@4205 | 2402 | // c_rarg1 - destination byte array address |
kvn@4205 | 2403 | // c_rarg2 - K (key) in little endian int array |
kvn@4205 | 2404 | // c_rarg3 - r vector byte array address |
kvn@4205 | 2405 | // c_rarg4 - input length |
kvn@4205 | 2406 | // |
kvn@6312 | 2407 | // Output: |
kvn@6312 | 2408 | // rax - input length |
kvn@6312 | 2409 | // |
kvn@4205 | 2410 | address generate_cipherBlockChaining_encryptAESCrypt() { |
kvn@4363 | 2411 | assert(UseAES, "need AES instructions and misaligned SSE support"); |
kvn@4205 | 2412 | __ align(CodeEntryAlignment); |
kvn@4205 | 2413 | StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt"); |
kvn@4205 | 2414 | address start = __ pc(); |
kvn@4205 | 2415 | |
kvn@4205 | 2416 | Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256; |
kvn@4205 | 2417 | const Register from = rsi; // source array address |
kvn@4205 | 2418 | const Register to = rdx; // destination array address |
kvn@4205 | 2419 | const Register key = rcx; // key array address |
kvn@4205 | 2420 | const Register rvec = rdi; // r byte array initialized from initvector array address |
kvn@4205 | 2421 | // and left with the results of the last encryption block |
kvn@4205 | 2422 | const Register len_reg = rbx; // src len (must be multiple of blocksize 16) |
kvn@4205 | 2423 | const Register pos = rax; |
kvn@4205 | 2424 | |
kvn@4205 | 2425 | // xmm register assignments for the loops below |
kvn@4205 | 2426 | const XMMRegister xmm_result = xmm0; |
kvn@4205 | 2427 | const XMMRegister xmm_temp = xmm1; |
kvn@4205 | 2428 | // first 6 keys preloaded into xmm2-xmm7 |
kvn@4205 | 2429 | const int XMM_REG_NUM_KEY_FIRST = 2; |
kvn@4205 | 2430 | const int XMM_REG_NUM_KEY_LAST = 7; |
kvn@4205 | 2431 | const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST); |
kvn@4205 | 2432 | |
kvn@4205 | 2433 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 2434 | handleSOERegisters(true /*saving*/); |
kvn@4205 | 2435 | |
kvn@4205 | 2436 | // load registers from incoming parameters |
kvn@4205 | 2437 | const Address from_param(rbp, 8+0); |
kvn@4205 | 2438 | const Address to_param (rbp, 8+4); |
kvn@4205 | 2439 | const Address key_param (rbp, 8+8); |
kvn@4205 | 2440 | const Address rvec_param (rbp, 8+12); |
kvn@4205 | 2441 | const Address len_param (rbp, 8+16); |
kvn@4205 | 2442 | __ movptr(from , from_param); |
kvn@4205 | 2443 | __ movptr(to , to_param); |
kvn@4205 | 2444 | __ movptr(key , key_param); |
kvn@4205 | 2445 | __ movptr(rvec , rvec_param); |
kvn@4205 | 2446 | __ movptr(len_reg , len_param); |
kvn@4205 | 2447 | |
kvn@4205 | 2448 | const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front |
kvn@4205 | 2449 | __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); |
kvn@4205 | 2450 | // load up xmm regs 2 thru 7 with keys 0-5 |
kvn@4205 | 2451 | for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { |
kvn@4205 | 2452 | load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask); |
kvn@4205 | 2453 | offset += 0x10; |
kvn@4205 | 2454 | } |
kvn@4205 | 2455 | |
kvn@4205 | 2456 | __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec |
kvn@4205 | 2457 | |
kvn@4205 | 2458 | // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256)) |
kvn@4205 | 2459 | __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); |
kvn@4205 | 2460 | __ cmpl(rax, 44); |
kvn@4205 | 2461 | __ jcc(Assembler::notEqual, L_key_192_256); |
kvn@4205 | 2462 | |
kvn@4205 | 2463 | // 128 bit code follows here |
kvn@4363 | 2464 | __ movl(pos, 0); |
kvn@4205 | 2465 | __ align(OptoLoopAlignment); |
kvn@4205 | 2466 | __ BIND(L_loopTop_128); |
kvn@4205 | 2467 | __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input |
kvn@4205 | 2468 | __ pxor (xmm_result, xmm_temp); // xor with the current r vector |
kvn@4205 | 2469 | |
kvn@4205 | 2470 | __ pxor (xmm_result, xmm_key0); // do the aes rounds |
kvn@4205 | 2471 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { |
kvn@4205 | 2472 | __ aesenc(xmm_result, as_XMMRegister(rnum)); |
kvn@4205 | 2473 | } |
kvn@4205 | 2474 | for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) { |
kvn@4205 | 2475 | aes_enc_key(xmm_result, xmm_temp, key, key_offset); |
kvn@4205 | 2476 | } |
kvn@4205 | 2477 | load_key(xmm_temp, key, 0xa0); |
kvn@4205 | 2478 | __ aesenclast(xmm_result, xmm_temp); |
kvn@4205 | 2479 | |
kvn@4205 | 2480 | __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output |
kvn@4205 | 2481 | // no need to store r to memory until we exit |
kvn@4205 | 2482 | __ addptr(pos, AESBlockSize); |
kvn@4205 | 2483 | __ subptr(len_reg, AESBlockSize); |
kvn@4205 | 2484 | __ jcc(Assembler::notEqual, L_loopTop_128); |
kvn@4205 | 2485 | |
kvn@4205 | 2486 | __ BIND(L_exit); |
kvn@4205 | 2487 | __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object |
kvn@4205 | 2488 | |
kvn@4205 | 2489 | handleSOERegisters(false /*restoring*/); |
kvn@6312 | 2490 | __ movptr(rax, len_param); // return length |
kvn@4205 | 2491 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 2492 | __ ret(0); |
kvn@4205 | 2493 | |
kvn@4363 | 2494 | __ BIND(L_key_192_256); |
kvn@4363 | 2495 | // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256) |
kvn@4205 | 2496 | __ cmpl(rax, 52); |
kvn@4205 | 2497 | __ jcc(Assembler::notEqual, L_key_256); |
kvn@4205 | 2498 | |
kvn@4205 | 2499 | // 192-bit code follows here (could be changed to use more xmm registers) |
kvn@4363 | 2500 | __ movl(pos, 0); |
kvn@4363 | 2501 | __ align(OptoLoopAlignment); |
kvn@4363 | 2502 | __ BIND(L_loopTop_192); |
kvn@4205 | 2503 | __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input |
kvn@4205 | 2504 | __ pxor (xmm_result, xmm_temp); // xor with the current r vector |
kvn@4205 | 2505 | |
kvn@4205 | 2506 | __ pxor (xmm_result, xmm_key0); // do the aes rounds |
kvn@4205 | 2507 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { |
kvn@4205 | 2508 | __ aesenc(xmm_result, as_XMMRegister(rnum)); |
kvn@4205 | 2509 | } |
kvn@4205 | 2510 | for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) { |
kvn@4205 | 2511 | aes_enc_key(xmm_result, xmm_temp, key, key_offset); |
kvn@4205 | 2512 | } |
kvn@4205 | 2513 | load_key(xmm_temp, key, 0xc0); |
kvn@4205 | 2514 | __ aesenclast(xmm_result, xmm_temp); |
kvn@4205 | 2515 | |
kvn@4205 | 2516 | __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output |
kvn@4205 | 2517 | // no need to store r to memory until we exit |
kvn@4205 | 2518 | __ addptr(pos, AESBlockSize); |
kvn@4205 | 2519 | __ subptr(len_reg, AESBlockSize); |
kvn@4205 | 2520 | __ jcc(Assembler::notEqual, L_loopTop_192); |
kvn@4205 | 2521 | __ jmp(L_exit); |
kvn@4205 | 2522 | |
kvn@4363 | 2523 | __ BIND(L_key_256); |
kvn@4205 | 2524 | // 256-bit code follows here (could be changed to use more xmm registers) |
kvn@4363 | 2525 | __ movl(pos, 0); |
kvn@4363 | 2526 | __ align(OptoLoopAlignment); |
kvn@4363 | 2527 | __ BIND(L_loopTop_256); |
kvn@4205 | 2528 | __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input |
kvn@4205 | 2529 | __ pxor (xmm_result, xmm_temp); // xor with the current r vector |
kvn@4205 | 2530 | |
kvn@4205 | 2531 | __ pxor (xmm_result, xmm_key0); // do the aes rounds |
kvn@4205 | 2532 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { |
kvn@4205 | 2533 | __ aesenc(xmm_result, as_XMMRegister(rnum)); |
kvn@4205 | 2534 | } |
kvn@4205 | 2535 | for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) { |
kvn@4205 | 2536 | aes_enc_key(xmm_result, xmm_temp, key, key_offset); |
kvn@4205 | 2537 | } |
kvn@4205 | 2538 | load_key(xmm_temp, key, 0xe0); |
kvn@4205 | 2539 | __ aesenclast(xmm_result, xmm_temp); |
kvn@4205 | 2540 | |
kvn@4205 | 2541 | __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output |
kvn@4205 | 2542 | // no need to store r to memory until we exit |
kvn@4205 | 2543 | __ addptr(pos, AESBlockSize); |
kvn@4205 | 2544 | __ subptr(len_reg, AESBlockSize); |
kvn@4205 | 2545 | __ jcc(Assembler::notEqual, L_loopTop_256); |
kvn@4205 | 2546 | __ jmp(L_exit); |
kvn@4205 | 2547 | |
kvn@4205 | 2548 | return start; |
kvn@4205 | 2549 | } |
kvn@4205 | 2550 | |
kvn@4205 | 2551 | |
kvn@4205 | 2552 | // CBC AES Decryption. |
kvn@4205 | 2553 | // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time. |
kvn@4205 | 2554 | // |
kvn@4205 | 2555 | // Arguments: |
kvn@4205 | 2556 | // |
kvn@4205 | 2557 | // Inputs: |
kvn@4205 | 2558 | // c_rarg0 - source byte array address |
kvn@4205 | 2559 | // c_rarg1 - destination byte array address |
kvn@4205 | 2560 | // c_rarg2 - K (key) in little endian int array |
kvn@4205 | 2561 | // c_rarg3 - r vector byte array address |
kvn@4205 | 2562 | // c_rarg4 - input length |
kvn@4205 | 2563 | // |
kvn@6312 | 2564 | // Output: |
kvn@6312 | 2565 | // rax - input length |
kvn@6312 | 2566 | // |
kvn@4205 | 2567 | |
kvn@4205 | 2568 | address generate_cipherBlockChaining_decryptAESCrypt() { |
kvn@4363 | 2569 | assert(UseAES, "need AES instructions and misaligned SSE support"); |
kvn@4205 | 2570 | __ align(CodeEntryAlignment); |
kvn@4205 | 2571 | StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt"); |
kvn@4205 | 2572 | address start = __ pc(); |
kvn@4205 | 2573 | |
kvn@4205 | 2574 | Label L_exit, L_key_192_256, L_key_256; |
kvn@4205 | 2575 | Label L_singleBlock_loopTop_128; |
kvn@4205 | 2576 | Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256; |
kvn@4205 | 2577 | const Register from = rsi; // source array address |
kvn@4205 | 2578 | const Register to = rdx; // destination array address |
kvn@4205 | 2579 | const Register key = rcx; // key array address |
kvn@4205 | 2580 | const Register rvec = rdi; // r byte array initialized from initvector array address |
kvn@4205 | 2581 | // and left with the results of the last encryption block |
kvn@4205 | 2582 | const Register len_reg = rbx; // src len (must be multiple of blocksize 16) |
kvn@4205 | 2583 | const Register pos = rax; |
kvn@4205 | 2584 | |
kvn@4205 | 2585 | // xmm register assignments for the loops below |
kvn@4205 | 2586 | const XMMRegister xmm_result = xmm0; |
kvn@4205 | 2587 | const XMMRegister xmm_temp = xmm1; |
kvn@4205 | 2588 | // first 6 keys preloaded into xmm2-xmm7 |
kvn@4205 | 2589 | const int XMM_REG_NUM_KEY_FIRST = 2; |
kvn@4205 | 2590 | const int XMM_REG_NUM_KEY_LAST = 7; |
kvn@4205 | 2591 | const int FIRST_NON_REG_KEY_offset = 0x70; |
kvn@4205 | 2592 | const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST); |
kvn@4205 | 2593 | |
kvn@4205 | 2594 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 2595 | handleSOERegisters(true /*saving*/); |
kvn@4205 | 2596 | |
kvn@4205 | 2597 | // load registers from incoming parameters |
kvn@4205 | 2598 | const Address from_param(rbp, 8+0); |
kvn@4205 | 2599 | const Address to_param (rbp, 8+4); |
kvn@4205 | 2600 | const Address key_param (rbp, 8+8); |
kvn@4205 | 2601 | const Address rvec_param (rbp, 8+12); |
kvn@4205 | 2602 | const Address len_param (rbp, 8+16); |
kvn@4205 | 2603 | __ movptr(from , from_param); |
kvn@4205 | 2604 | __ movptr(to , to_param); |
kvn@4205 | 2605 | __ movptr(key , key_param); |
kvn@4205 | 2606 | __ movptr(rvec , rvec_param); |
kvn@4205 | 2607 | __ movptr(len_reg , len_param); |
kvn@4205 | 2608 | |
kvn@4205 | 2609 | // the java expanded key ordering is rotated one position from what we want |
kvn@4205 | 2610 | // so we start from 0x10 here and hit 0x00 last |
kvn@4205 | 2611 | const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front |
kvn@4205 | 2612 | __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); |
kvn@4205 | 2613 | // load up xmm regs 2 thru 6 with first 5 keys |
kvn@4205 | 2614 | for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { |
kvn@4205 | 2615 | load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask); |
kvn@4205 | 2616 | offset += 0x10; |
kvn@4205 | 2617 | } |
kvn@4205 | 2618 | |
kvn@4205 | 2619 | // inside here, use the rvec register to point to previous block cipher |
kvn@4205 | 2620 | // with which we xor at the end of each newly decrypted block |
kvn@4205 | 2621 | const Register prev_block_cipher_ptr = rvec; |
kvn@4205 | 2622 | |
kvn@4205 | 2623 | // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256)) |
kvn@4205 | 2624 | __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); |
kvn@4205 | 2625 | __ cmpl(rax, 44); |
kvn@4205 | 2626 | __ jcc(Assembler::notEqual, L_key_192_256); |
kvn@4205 | 2627 | |
kvn@4205 | 2628 | |
kvn@4205 | 2629 | // 128-bit code follows here, parallelized |
kvn@4363 | 2630 | __ movl(pos, 0); |
kvn@4363 | 2631 | __ align(OptoLoopAlignment); |
kvn@4363 | 2632 | __ BIND(L_singleBlock_loopTop_128); |
kvn@4205 | 2633 | __ cmpptr(len_reg, 0); // any blocks left?? |
kvn@4205 | 2634 | __ jcc(Assembler::equal, L_exit); |
kvn@4205 | 2635 | __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input |
kvn@4205 | 2636 | __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds |
kvn@4205 | 2637 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { |
kvn@4205 | 2638 | __ aesdec(xmm_result, as_XMMRegister(rnum)); |
kvn@4205 | 2639 | } |
kvn@4205 | 2640 | for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xa0; key_offset += 0x10) { // 128-bit runs up to key offset a0 |
kvn@4205 | 2641 | aes_dec_key(xmm_result, xmm_temp, key, key_offset); |
kvn@4205 | 2642 | } |
kvn@4205 | 2643 | load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0 |
kvn@4205 | 2644 | __ aesdeclast(xmm_result, xmm_temp); |
kvn@4205 | 2645 | __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00)); |
kvn@4205 | 2646 | __ pxor (xmm_result, xmm_temp); // xor with the current r vector |
kvn@4205 | 2647 | __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output |
kvn@4205 | 2648 | // no need to store r to memory until we exit |
kvn@4205 | 2649 | __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr |
kvn@4205 | 2650 | __ addptr(pos, AESBlockSize); |
kvn@4205 | 2651 | __ subptr(len_reg, AESBlockSize); |
kvn@4205 | 2652 | __ jmp(L_singleBlock_loopTop_128); |
kvn@4205 | 2653 | |
kvn@4205 | 2654 | |
kvn@4205 | 2655 | __ BIND(L_exit); |
kvn@4205 | 2656 | __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00)); |
kvn@4205 | 2657 | __ movptr(rvec , rvec_param); // restore this since used in loop |
kvn@4205 | 2658 | __ movdqu(Address(rvec, 0), xmm_temp); // final value of r stored in rvec of CipherBlockChaining object |
kvn@4205 | 2659 | handleSOERegisters(false /*restoring*/); |
kvn@6312 | 2660 | __ movptr(rax, len_param); // return length |
kvn@4205 | 2661 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
kvn@4205 | 2662 | __ ret(0); |
kvn@4205 | 2663 | |
kvn@4205 | 2664 | |
kvn@4205 | 2665 | __ BIND(L_key_192_256); |
kvn@4205 | 2666 | // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256) |
kvn@4205 | 2667 | __ cmpl(rax, 52); |
kvn@4205 | 2668 | __ jcc(Assembler::notEqual, L_key_256); |
kvn@4205 | 2669 | |
kvn@4205 | 2670 | // 192-bit code follows here (could be optimized to use parallelism) |
kvn@4363 | 2671 | __ movl(pos, 0); |
kvn@4205 | 2672 | __ align(OptoLoopAlignment); |
kvn@4205 | 2673 | __ BIND(L_singleBlock_loopTop_192); |
kvn@4205 | 2674 | __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input |
kvn@4205 | 2675 | __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds |
kvn@4205 | 2676 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { |
kvn@4205 | 2677 | __ aesdec(xmm_result, as_XMMRegister(rnum)); |
kvn@4205 | 2678 | } |
kvn@4205 | 2679 | for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xc0; key_offset += 0x10) { // 192-bit runs up to key offset c0 |
kvn@4205 | 2680 | aes_dec_key(xmm_result, xmm_temp, key, key_offset); |
kvn@4205 | 2681 | } |
kvn@4205 | 2682 | load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0 |
kvn@4205 | 2683 | __ aesdeclast(xmm_result, xmm_temp); |
kvn@4205 | 2684 | __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00)); |
kvn@4205 | 2685 | __ pxor (xmm_result, xmm_temp); // xor with the current r vector |
kvn@4205 | 2686 | __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output |
kvn@4205 | 2687 | // no need to store r to memory until we exit |
kvn@4205 | 2688 | __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr |
kvn@4205 | 2689 | __ addptr(pos, AESBlockSize); |
kvn@4205 | 2690 | __ subptr(len_reg, AESBlockSize); |
kvn@4205 | 2691 | __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192); |
kvn@4205 | 2692 | __ jmp(L_exit); |
kvn@4205 | 2693 | |
kvn@4205 | 2694 | __ BIND(L_key_256); |
kvn@4205 | 2695 | // 256-bit code follows here (could be optimized to use parallelism) |
kvn@4363 | 2696 | __ movl(pos, 0); |
kvn@4205 | 2697 | __ align(OptoLoopAlignment); |
kvn@4205 | 2698 | __ BIND(L_singleBlock_loopTop_256); |
kvn@4205 | 2699 | __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input |
kvn@4205 | 2700 | __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds |
kvn@4205 | 2701 | for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { |
kvn@4205 | 2702 | __ aesdec(xmm_result, as_XMMRegister(rnum)); |
kvn@4205 | 2703 | } |
kvn@4205 | 2704 | for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xe0; key_offset += 0x10) { // 256-bit runs up to key offset e0 |
kvn@4205 | 2705 | aes_dec_key(xmm_result, xmm_temp, key, key_offset); |
kvn@4205 | 2706 | } |
kvn@4205 | 2707 | load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0 |
kvn@4205 | 2708 | __ aesdeclast(xmm_result, xmm_temp); |
kvn@4205 | 2709 | __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00)); |
kvn@4205 | 2710 | __ pxor (xmm_result, xmm_temp); // xor with the current r vector |
kvn@4205 | 2711 | __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output |
kvn@4205 | 2712 | // no need to store r to memory until we exit |
kvn@4205 | 2713 | __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr |
kvn@4205 | 2714 | __ addptr(pos, AESBlockSize); |
kvn@4205 | 2715 | __ subptr(len_reg, AESBlockSize); |
kvn@4205 | 2716 | __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256); |
kvn@4205 | 2717 | __ jmp(L_exit); |
kvn@4205 | 2718 | |
kvn@4205 | 2719 | return start; |
kvn@4205 | 2720 | } |
kvn@4205 | 2721 | |
ascarpino@9788 | 2722 | // byte swap x86 long |
ascarpino@9788 | 2723 | address generate_ghash_long_swap_mask() { |
ascarpino@9788 | 2724 | __ align(CodeEntryAlignment); |
ascarpino@9788 | 2725 | StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask"); |
ascarpino@9788 | 2726 | address start = __ pc(); |
ascarpino@9788 | 2727 | __ emit_data(0x0b0a0908, relocInfo::none, 0); |
ascarpino@9788 | 2728 | __ emit_data(0x0f0e0d0c, relocInfo::none, 0); |
ascarpino@9788 | 2729 | __ emit_data(0x03020100, relocInfo::none, 0); |
ascarpino@9788 | 2730 | __ emit_data(0x07060504, relocInfo::none, 0); |
ascarpino@9788 | 2731 | |
ascarpino@9788 | 2732 | return start; |
ascarpino@9788 | 2733 | } |
ascarpino@9788 | 2734 | |
ascarpino@9788 | 2735 | // byte swap x86 byte array |
ascarpino@9788 | 2736 | address generate_ghash_byte_swap_mask() { |
ascarpino@9788 | 2737 | __ align(CodeEntryAlignment); |
ascarpino@9788 | 2738 | StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask"); |
ascarpino@9788 | 2739 | address start = __ pc(); |
ascarpino@9788 | 2740 | __ emit_data(0x0c0d0e0f, relocInfo::none, 0); |
ascarpino@9788 | 2741 | __ emit_data(0x08090a0b, relocInfo::none, 0); |
ascarpino@9788 | 2742 | __ emit_data(0x04050607, relocInfo::none, 0); |
ascarpino@9788 | 2743 | __ emit_data(0x00010203, relocInfo::none, 0); |
ascarpino@9788 | 2744 | return start; |
ascarpino@9788 | 2745 | } |
ascarpino@9788 | 2746 | |
ascarpino@9788 | 2747 | /* Single and multi-block ghash operations */ |
ascarpino@9788 | 2748 | address generate_ghash_processBlocks() { |
ascarpino@9788 | 2749 | assert(UseGHASHIntrinsics, "need GHASH intrinsics and CLMUL support"); |
ascarpino@9788 | 2750 | __ align(CodeEntryAlignment); |
ascarpino@9788 | 2751 | Label L_ghash_loop, L_exit; |
ascarpino@9788 | 2752 | StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks"); |
ascarpino@9788 | 2753 | address start = __ pc(); |
ascarpino@9788 | 2754 | |
ascarpino@9788 | 2755 | const Register state = rdi; |
ascarpino@9788 | 2756 | const Register subkeyH = rsi; |
ascarpino@9788 | 2757 | const Register data = rdx; |
ascarpino@9788 | 2758 | const Register blocks = rcx; |
ascarpino@9788 | 2759 | |
ascarpino@9788 | 2760 | const Address state_param(rbp, 8+0); |
ascarpino@9788 | 2761 | const Address subkeyH_param(rbp, 8+4); |
ascarpino@9788 | 2762 | const Address data_param(rbp, 8+8); |
ascarpino@9788 | 2763 | const Address blocks_param(rbp, 8+12); |
ascarpino@9788 | 2764 | |
ascarpino@9788 | 2765 | const XMMRegister xmm_temp0 = xmm0; |
ascarpino@9788 | 2766 | const XMMRegister xmm_temp1 = xmm1; |
ascarpino@9788 | 2767 | const XMMRegister xmm_temp2 = xmm2; |
ascarpino@9788 | 2768 | const XMMRegister xmm_temp3 = xmm3; |
ascarpino@9788 | 2769 | const XMMRegister xmm_temp4 = xmm4; |
ascarpino@9788 | 2770 | const XMMRegister xmm_temp5 = xmm5; |
ascarpino@9788 | 2771 | const XMMRegister xmm_temp6 = xmm6; |
ascarpino@9788 | 2772 | const XMMRegister xmm_temp7 = xmm7; |
ascarpino@9788 | 2773 | |
ascarpino@9788 | 2774 | __ enter(); |
ascarpino@9789 | 2775 | handleSOERegisters(true); // Save registers |
ascarpino@9788 | 2776 | |
ascarpino@9788 | 2777 | __ movptr(state, state_param); |
ascarpino@9788 | 2778 | __ movptr(subkeyH, subkeyH_param); |
ascarpino@9788 | 2779 | __ movptr(data, data_param); |
ascarpino@9788 | 2780 | __ movptr(blocks, blocks_param); |
ascarpino@9788 | 2781 | |
ascarpino@9788 | 2782 | __ movdqu(xmm_temp0, Address(state, 0)); |
ascarpino@9788 | 2783 | __ pshufb(xmm_temp0, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr())); |
ascarpino@9788 | 2784 | |
ascarpino@9788 | 2785 | __ movdqu(xmm_temp1, Address(subkeyH, 0)); |
ascarpino@9788 | 2786 | __ pshufb(xmm_temp1, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr())); |
ascarpino@9788 | 2787 | |
ascarpino@9788 | 2788 | __ BIND(L_ghash_loop); |
ascarpino@9788 | 2789 | __ movdqu(xmm_temp2, Address(data, 0)); |
ascarpino@9788 | 2790 | __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr())); |
ascarpino@9788 | 2791 | |
ascarpino@9788 | 2792 | __ pxor(xmm_temp0, xmm_temp2); |
ascarpino@9788 | 2793 | |
ascarpino@9788 | 2794 | // |
ascarpino@9788 | 2795 | // Multiply with the hash key |
ascarpino@9788 | 2796 | // |
ascarpino@9788 | 2797 | __ movdqu(xmm_temp3, xmm_temp0); |
ascarpino@9788 | 2798 | __ pclmulqdq(xmm_temp3, xmm_temp1, 0); // xmm3 holds a0*b0 |
ascarpino@9788 | 2799 | __ movdqu(xmm_temp4, xmm_temp0); |
ascarpino@9788 | 2800 | __ pclmulqdq(xmm_temp4, xmm_temp1, 16); // xmm4 holds a0*b1 |
ascarpino@9788 | 2801 | |
ascarpino@9788 | 2802 | __ movdqu(xmm_temp5, xmm_temp0); |
ascarpino@9788 | 2803 | __ pclmulqdq(xmm_temp5, xmm_temp1, 1); // xmm5 holds a1*b0 |
ascarpino@9788 | 2804 | __ movdqu(xmm_temp6, xmm_temp0); |
ascarpino@9788 | 2805 | __ pclmulqdq(xmm_temp6, xmm_temp1, 17); // xmm6 holds a1*b1 |
ascarpino@9788 | 2806 | |
ascarpino@9788 | 2807 | __ pxor(xmm_temp4, xmm_temp5); // xmm4 holds a0*b1 + a1*b0 |
ascarpino@9788 | 2808 | |
ascarpino@9788 | 2809 | __ movdqu(xmm_temp5, xmm_temp4); // move the contents of xmm4 to xmm5 |
ascarpino@9788 | 2810 | __ psrldq(xmm_temp4, 8); // shift by xmm4 64 bits to the right |
ascarpino@9788 | 2811 | __ pslldq(xmm_temp5, 8); // shift by xmm5 64 bits to the left |
ascarpino@9788 | 2812 | __ pxor(xmm_temp3, xmm_temp5); |
ascarpino@9788 | 2813 | __ pxor(xmm_temp6, xmm_temp4); // Register pair <xmm6:xmm3> holds the result |
ascarpino@9788 | 2814 | // of the carry-less multiplication of |
ascarpino@9788 | 2815 | // xmm0 by xmm1. |
ascarpino@9788 | 2816 | |
ascarpino@9788 | 2817 | // We shift the result of the multiplication by one bit position |
ascarpino@9788 | 2818 | // to the left to cope for the fact that the bits are reversed. |
ascarpino@9788 | 2819 | __ movdqu(xmm_temp7, xmm_temp3); |
ascarpino@9788 | 2820 | __ movdqu(xmm_temp4, xmm_temp6); |
ascarpino@9788 | 2821 | __ pslld (xmm_temp3, 1); |
ascarpino@9788 | 2822 | __ pslld(xmm_temp6, 1); |
ascarpino@9788 | 2823 | __ psrld(xmm_temp7, 31); |
ascarpino@9788 | 2824 | __ psrld(xmm_temp4, 31); |
ascarpino@9788 | 2825 | __ movdqu(xmm_temp5, xmm_temp7); |
ascarpino@9788 | 2826 | __ pslldq(xmm_temp4, 4); |
ascarpino@9788 | 2827 | __ pslldq(xmm_temp7, 4); |
ascarpino@9788 | 2828 | __ psrldq(xmm_temp5, 12); |
ascarpino@9788 | 2829 | __ por(xmm_temp3, xmm_temp7); |
ascarpino@9788 | 2830 | __ por(xmm_temp6, xmm_temp4); |
ascarpino@9788 | 2831 | __ por(xmm_temp6, xmm_temp5); |
ascarpino@9788 | 2832 | |
ascarpino@9788 | 2833 | // |
ascarpino@9788 | 2834 | // First phase of the reduction |
ascarpino@9788 | 2835 | // |
ascarpino@9788 | 2836 | // Move xmm3 into xmm4, xmm5, xmm7 in order to perform the shifts |
ascarpino@9788 | 2837 | // independently. |
ascarpino@9788 | 2838 | __ movdqu(xmm_temp7, xmm_temp3); |
ascarpino@9788 | 2839 | __ movdqu(xmm_temp4, xmm_temp3); |
ascarpino@9788 | 2840 | __ movdqu(xmm_temp5, xmm_temp3); |
ascarpino@9788 | 2841 | __ pslld(xmm_temp7, 31); // packed right shift shifting << 31 |
ascarpino@9788 | 2842 | __ pslld(xmm_temp4, 30); // packed right shift shifting << 30 |
ascarpino@9788 | 2843 | __ pslld(xmm_temp5, 25); // packed right shift shifting << 25 |
ascarpino@9788 | 2844 | __ pxor(xmm_temp7, xmm_temp4); // xor the shifted versions |
ascarpino@9788 | 2845 | __ pxor(xmm_temp7, xmm_temp5); |
ascarpino@9788 | 2846 | __ movdqu(xmm_temp4, xmm_temp7); |
ascarpino@9788 | 2847 | __ pslldq(xmm_temp7, 12); |
ascarpino@9788 | 2848 | __ psrldq(xmm_temp4, 4); |
ascarpino@9788 | 2849 | __ pxor(xmm_temp3, xmm_temp7); // first phase of the reduction complete |
ascarpino@9788 | 2850 | |
ascarpino@9788 | 2851 | // |
ascarpino@9788 | 2852 | // Second phase of the reduction |
ascarpino@9788 | 2853 | // |
ascarpino@9788 | 2854 | // Make 3 copies of xmm3 in xmm2, xmm5, xmm7 for doing these |
ascarpino@9788 | 2855 | // shift operations. |
ascarpino@9788 | 2856 | __ movdqu(xmm_temp2, xmm_temp3); |
ascarpino@9788 | 2857 | __ movdqu(xmm_temp7, xmm_temp3); |
ascarpino@9788 | 2858 | __ movdqu(xmm_temp5, xmm_temp3); |
ascarpino@9788 | 2859 | __ psrld(xmm_temp2, 1); // packed left shifting >> 1 |
ascarpino@9788 | 2860 | __ psrld(xmm_temp7, 2); // packed left shifting >> 2 |
ascarpino@9788 | 2861 | __ psrld(xmm_temp5, 7); // packed left shifting >> 7 |
ascarpino@9788 | 2862 | __ pxor(xmm_temp2, xmm_temp7); // xor the shifted versions |
ascarpino@9788 | 2863 | __ pxor(xmm_temp2, xmm_temp5); |
ascarpino@9788 | 2864 | __ pxor(xmm_temp2, xmm_temp4); |
ascarpino@9788 | 2865 | __ pxor(xmm_temp3, xmm_temp2); |
ascarpino@9788 | 2866 | __ pxor(xmm_temp6, xmm_temp3); // the result is in xmm6 |
ascarpino@9788 | 2867 | |
ascarpino@9788 | 2868 | __ decrement(blocks); |
ascarpino@9788 | 2869 | __ jcc(Assembler::zero, L_exit); |
ascarpino@9788 | 2870 | __ movdqu(xmm_temp0, xmm_temp6); |
ascarpino@9788 | 2871 | __ addptr(data, 16); |
ascarpino@9788 | 2872 | __ jmp(L_ghash_loop); |
ascarpino@9788 | 2873 | |
ascarpino@9788 | 2874 | __ BIND(L_exit); |
ascarpino@9788 | 2875 | // Byte swap 16-byte result |
ascarpino@9788 | 2876 | __ pshufb(xmm_temp6, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr())); |
ascarpino@9788 | 2877 | __ movdqu(Address(state, 0), xmm_temp6); // store the result |
ascarpino@9788 | 2878 | |
ascarpino@9789 | 2879 | handleSOERegisters(false); // restore registers |
ascarpino@9788 | 2880 | __ leave(); |
ascarpino@9788 | 2881 | __ ret(0); |
ascarpino@9788 | 2882 | return start; |
ascarpino@9788 | 2883 | } |
ascarpino@9788 | 2884 | |
drchase@5353 | 2885 | /** |
drchase@5353 | 2886 | * Arguments: |
drchase@5353 | 2887 | * |
drchase@5353 | 2888 | * Inputs: |
drchase@5353 | 2889 | * rsp(4) - int crc |
drchase@5353 | 2890 | * rsp(8) - byte* buf |
drchase@5353 | 2891 | * rsp(12) - int length |
drchase@5353 | 2892 | * |
drchase@5353 | 2893 | * Ouput: |
drchase@5353 | 2894 | * rax - int crc result |
drchase@5353 | 2895 | */ |
drchase@5353 | 2896 | address generate_updateBytesCRC32() { |
drchase@5353 | 2897 | assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions"); |
drchase@5353 | 2898 | |
drchase@5353 | 2899 | __ align(CodeEntryAlignment); |
drchase@5353 | 2900 | StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32"); |
drchase@5353 | 2901 | |
drchase@5353 | 2902 | address start = __ pc(); |
drchase@5353 | 2903 | |
drchase@5353 | 2904 | const Register crc = rdx; // crc |
drchase@5353 | 2905 | const Register buf = rsi; // source java byte array address |
drchase@5353 | 2906 | const Register len = rcx; // length |
drchase@5353 | 2907 | const Register table = rdi; // crc_table address (reuse register) |
drchase@5353 | 2908 | const Register tmp = rbx; |
drchase@5353 | 2909 | assert_different_registers(crc, buf, len, table, tmp, rax); |
drchase@5353 | 2910 | |
drchase@5353 | 2911 | BLOCK_COMMENT("Entry:"); |
drchase@5353 | 2912 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
drchase@5353 | 2913 | __ push(rsi); |
drchase@5353 | 2914 | __ push(rdi); |
drchase@5353 | 2915 | __ push(rbx); |
drchase@5353 | 2916 | |
drchase@5353 | 2917 | Address crc_arg(rbp, 8 + 0); |
drchase@5353 | 2918 | Address buf_arg(rbp, 8 + 4); |
drchase@5353 | 2919 | Address len_arg(rbp, 8 + 8); |
drchase@5353 | 2920 | |
drchase@5353 | 2921 | // Load up: |
drchase@5353 | 2922 | __ movl(crc, crc_arg); |
drchase@5353 | 2923 | __ movptr(buf, buf_arg); |
drchase@5353 | 2924 | __ movl(len, len_arg); |
drchase@5353 | 2925 | |
drchase@5353 | 2926 | __ kernel_crc32(crc, buf, len, table, tmp); |
drchase@5353 | 2927 | |
drchase@5353 | 2928 | __ movl(rax, crc); |
drchase@5353 | 2929 | __ pop(rbx); |
drchase@5353 | 2930 | __ pop(rdi); |
drchase@5353 | 2931 | __ pop(rsi); |
drchase@5353 | 2932 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
drchase@5353 | 2933 | __ ret(0); |
drchase@5353 | 2934 | |
drchase@5353 | 2935 | return start; |
drchase@5353 | 2936 | } |
drchase@5353 | 2937 | |
goetz@5400 | 2938 | // Safefetch stubs. |
goetz@5400 | 2939 | void generate_safefetch(const char* name, int size, address* entry, |
goetz@5400 | 2940 | address* fault_pc, address* continuation_pc) { |
goetz@5400 | 2941 | // safefetch signatures: |
goetz@5400 | 2942 | // int SafeFetch32(int* adr, int errValue); |
goetz@5400 | 2943 | // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue); |
goetz@5400 | 2944 | |
goetz@5400 | 2945 | StubCodeMark mark(this, "StubRoutines", name); |
goetz@5400 | 2946 | |
goetz@5400 | 2947 | // Entry point, pc or function descriptor. |
goetz@5400 | 2948 | *entry = __ pc(); |
goetz@5400 | 2949 | |
goetz@5400 | 2950 | __ movl(rax, Address(rsp, 0x8)); |
goetz@5400 | 2951 | __ movl(rcx, Address(rsp, 0x4)); |
goetz@5400 | 2952 | // Load *adr into eax, may fault. |
goetz@5400 | 2953 | *fault_pc = __ pc(); |
goetz@5400 | 2954 | switch (size) { |
goetz@5400 | 2955 | case 4: |
goetz@5400 | 2956 | // int32_t |
goetz@5400 | 2957 | __ movl(rax, Address(rcx, 0)); |
goetz@5400 | 2958 | break; |
goetz@5400 | 2959 | case 8: |
goetz@5400 | 2960 | // int64_t |
goetz@5400 | 2961 | Unimplemented(); |
goetz@5400 | 2962 | break; |
goetz@5400 | 2963 | default: |
goetz@5400 | 2964 | ShouldNotReachHere(); |
goetz@5400 | 2965 | } |
goetz@5400 | 2966 | |
goetz@5400 | 2967 | // Return errValue or *adr. |
goetz@5400 | 2968 | *continuation_pc = __ pc(); |
goetz@5400 | 2969 | __ ret(0); |
goetz@5400 | 2970 | } |
kvn@4205 | 2971 | |
duke@435 | 2972 | public: |
duke@435 | 2973 | // Information about frame layout at time of blocking runtime call. |
duke@435 | 2974 | // Note that we only have to preserve callee-saved registers since |
duke@435 | 2975 | // the compilers are responsible for supplying a continuation point |
duke@435 | 2976 | // if they expect all registers to be preserved. |
duke@435 | 2977 | enum layout { |
duke@435 | 2978 | thread_off, // last_java_sp |
never@2978 | 2979 | arg1_off, |
never@2978 | 2980 | arg2_off, |
duke@435 | 2981 | rbp_off, // callee saved register |
duke@435 | 2982 | ret_pc, |
duke@435 | 2983 | framesize |
duke@435 | 2984 | }; |
duke@435 | 2985 | |
duke@435 | 2986 | private: |
duke@435 | 2987 | |
duke@435 | 2988 | #undef __ |
duke@435 | 2989 | #define __ masm-> |
duke@435 | 2990 | |
duke@435 | 2991 | //------------------------------------------------------------------------------------------------------------------------ |
duke@435 | 2992 | // Continuation point for throwing of implicit exceptions that are not handled in |
duke@435 | 2993 | // the current activation. Fabricates an exception oop and initiates normal |
duke@435 | 2994 | // exception dispatching in this frame. |
duke@435 | 2995 | // |
duke@435 | 2996 | // Previously the compiler (c2) allowed for callee save registers on Java calls. |
duke@435 | 2997 | // This is no longer true after adapter frames were removed but could possibly |
duke@435 | 2998 | // be brought back in the future if the interpreter code was reworked and it |
duke@435 | 2999 | // was deemed worthwhile. The comment below was left to describe what must |
duke@435 | 3000 | // happen here if callee saves were resurrected. As it stands now this stub |
duke@435 | 3001 | // could actually be a vanilla BufferBlob and have now oopMap at all. |
duke@435 | 3002 | // Since it doesn't make much difference we've chosen to leave it the |
duke@435 | 3003 | // way it was in the callee save days and keep the comment. |
duke@435 | 3004 | |
duke@435 | 3005 | // If we need to preserve callee-saved values we need a callee-saved oop map and |
duke@435 | 3006 | // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs. |
duke@435 | 3007 | // If the compiler needs all registers to be preserved between the fault |
duke@435 | 3008 | // point and the exception handler then it must assume responsibility for that in |
duke@435 | 3009 | // AbstractCompiler::continuation_for_implicit_null_exception or |
duke@435 | 3010 | // continuation_for_implicit_division_by_zero_exception. All other implicit |
duke@435 | 3011 | // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are |
duke@435 | 3012 | // either at call sites or otherwise assume that stack unwinding will be initiated, |
duke@435 | 3013 | // so caller saved registers were assumed volatile in the compiler. |
duke@435 | 3014 | address generate_throw_exception(const char* name, address runtime_entry, |
never@3136 | 3015 | Register arg1 = noreg, Register arg2 = noreg) { |
duke@435 | 3016 | |
duke@435 | 3017 | int insts_size = 256; |
duke@435 | 3018 | int locs_size = 32; |
duke@435 | 3019 | |
duke@435 | 3020 | CodeBuffer code(name, insts_size, locs_size); |
duke@435 | 3021 | OopMapSet* oop_maps = new OopMapSet(); |
duke@435 | 3022 | MacroAssembler* masm = new MacroAssembler(&code); |
duke@435 | 3023 | |
duke@435 | 3024 | address start = __ pc(); |
duke@435 | 3025 | |
duke@435 | 3026 | // This is an inlined and slightly modified version of call_VM |
duke@435 | 3027 | // which has the ability to fetch the return PC out of |
duke@435 | 3028 | // thread-local storage and also sets up last_Java_sp slightly |
duke@435 | 3029 | // differently than the real call_VM |
duke@435 | 3030 | Register java_thread = rbx; |
duke@435 | 3031 | __ get_thread(java_thread); |
duke@435 | 3032 | |
duke@435 | 3033 | __ enter(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 3034 | |
duke@435 | 3035 | // pc and rbp, already pushed |
never@739 | 3036 | __ subptr(rsp, (framesize-2) * wordSize); // prolog |
duke@435 | 3037 | |
duke@435 | 3038 | // Frame is now completed as far as size and linkage. |
duke@435 | 3039 | |
duke@435 | 3040 | int frame_complete = __ pc() - start; |
duke@435 | 3041 | |
duke@435 | 3042 | // push java thread (becomes first argument of C function) |
never@739 | 3043 | __ movptr(Address(rsp, thread_off * wordSize), java_thread); |
never@2978 | 3044 | if (arg1 != noreg) { |
never@2978 | 3045 | __ movptr(Address(rsp, arg1_off * wordSize), arg1); |
never@2978 | 3046 | } |
never@2978 | 3047 | if (arg2 != noreg) { |
never@2978 | 3048 | assert(arg1 != noreg, "missing reg arg"); |
never@2978 | 3049 | __ movptr(Address(rsp, arg2_off * wordSize), arg2); |
never@2978 | 3050 | } |
duke@435 | 3051 | |
duke@435 | 3052 | // Set up last_Java_sp and last_Java_fp |
duke@435 | 3053 | __ set_last_Java_frame(java_thread, rsp, rbp, NULL); |
duke@435 | 3054 | |
duke@435 | 3055 | // Call runtime |
duke@435 | 3056 | BLOCK_COMMENT("call runtime_entry"); |
duke@435 | 3057 | __ call(RuntimeAddress(runtime_entry)); |
duke@435 | 3058 | // Generate oop map |
duke@435 | 3059 | OopMap* map = new OopMap(framesize, 0); |
duke@435 | 3060 | oop_maps->add_gc_map(__ pc() - start, map); |
duke@435 | 3061 | |
duke@435 | 3062 | // restore the thread (cannot use the pushed argument since arguments |
duke@435 | 3063 | // may be overwritten by C code generated by an optimizing compiler); |
duke@435 | 3064 | // however can use the register value directly if it is callee saved. |
duke@435 | 3065 | __ get_thread(java_thread); |
duke@435 | 3066 | |
kevinw@8877 | 3067 | __ reset_last_Java_frame(java_thread, true); |
duke@435 | 3068 | |
duke@435 | 3069 | __ leave(); // required for proper stackwalking of RuntimeStub frame |
duke@435 | 3070 | |
duke@435 | 3071 | // check for pending exceptions |
duke@435 | 3072 | #ifdef ASSERT |
duke@435 | 3073 | Label L; |
never@739 | 3074 | __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
duke@435 | 3075 | __ jcc(Assembler::notEqual, L); |
duke@435 | 3076 | __ should_not_reach_here(); |
duke@435 | 3077 | __ bind(L); |
duke@435 | 3078 | #endif /* ASSERT */ |
duke@435 | 3079 | __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); |
duke@435 | 3080 | |
duke@435 | 3081 | |
duke@435 | 3082 | RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false); |
duke@435 | 3083 | return stub->entry_point(); |
duke@435 | 3084 | } |
duke@435 | 3085 | |
duke@435 | 3086 | |
duke@435 | 3087 | void create_control_words() { |
duke@435 | 3088 | // Round to nearest, 53-bit mode, exceptions masked |
duke@435 | 3089 | StubRoutines::_fpu_cntrl_wrd_std = 0x027F; |
duke@435 | 3090 | // Round to zero, 53-bit mode, exception mased |
duke@435 | 3091 | StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F; |
duke@435 | 3092 | // Round to nearest, 24-bit mode, exceptions masked |
duke@435 | 3093 | StubRoutines::_fpu_cntrl_wrd_24 = 0x007F; |
duke@435 | 3094 | // Round to nearest, 64-bit mode, exceptions masked |
duke@435 | 3095 | StubRoutines::_fpu_cntrl_wrd_64 = 0x037F; |
duke@435 | 3096 | // Round to nearest, 64-bit mode, exceptions masked |
duke@435 | 3097 | StubRoutines::_mxcsr_std = 0x1F80; |
duke@435 | 3098 | // Note: the following two constants are 80-bit values |
duke@435 | 3099 | // layout is critical for correct loading by FPU. |
duke@435 | 3100 | // Bias for strict fp multiply/divide |
duke@435 | 3101 | StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000 |
duke@435 | 3102 | StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000; |
duke@435 | 3103 | StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff; |
duke@435 | 3104 | // Un-Bias for strict fp multiply/divide |
duke@435 | 3105 | StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000 |
duke@435 | 3106 | StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000; |
duke@435 | 3107 | StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff; |
duke@435 | 3108 | } |
duke@435 | 3109 | |
duke@435 | 3110 | //--------------------------------------------------------------------------- |
duke@435 | 3111 | // Initialization |
duke@435 | 3112 | |
duke@435 | 3113 | void generate_initial() { |
duke@435 | 3114 | // Generates all stubs and initializes the entry points |
duke@435 | 3115 | |
duke@435 | 3116 | //------------------------------------------------------------------------------------------------------------------------ |
duke@435 | 3117 | // entry points that exist in all platforms |
duke@435 | 3118 | // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than |
duke@435 | 3119 | // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp. |
duke@435 | 3120 | StubRoutines::_forward_exception_entry = generate_forward_exception(); |
duke@435 | 3121 | |
duke@435 | 3122 | StubRoutines::_call_stub_entry = |
duke@435 | 3123 | generate_call_stub(StubRoutines::_call_stub_return_address); |
duke@435 | 3124 | // is referenced by megamorphic call |
duke@435 | 3125 | StubRoutines::_catch_exception_entry = generate_catch_exception(); |
duke@435 | 3126 | |
duke@435 | 3127 | // These are currently used by Solaris/Intel |
duke@435 | 3128 | StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); |
duke@435 | 3129 | |
duke@435 | 3130 | StubRoutines::_handler_for_unsafe_access_entry = |
duke@435 | 3131 | generate_handler_for_unsafe_access(); |
duke@435 | 3132 | |
duke@435 | 3133 | // platform dependent |
duke@435 | 3134 | create_control_words(); |
duke@435 | 3135 | |
never@739 | 3136 | StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr(); |
never@739 | 3137 | StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd(); |
duke@435 | 3138 | StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT, |
duke@435 | 3139 | CAST_FROM_FN_PTR(address, SharedRuntime::d2i)); |
duke@435 | 3140 | StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG, |
duke@435 | 3141 | CAST_FROM_FN_PTR(address, SharedRuntime::d2l)); |
never@2978 | 3142 | |
never@2978 | 3143 | // Build this early so it's available for the interpreter |
bdelsart@3372 | 3144 | StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError)); |
drchase@5353 | 3145 | |
drchase@5353 | 3146 | if (UseCRC32Intrinsics) { |
drchase@5353 | 3147 | // set table address before stub generation which use it |
drchase@5353 | 3148 | StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table; |
drchase@5353 | 3149 | StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32(); |
drchase@5353 | 3150 | } |
duke@435 | 3151 | } |
duke@435 | 3152 | |
duke@435 | 3153 | |
duke@435 | 3154 | void generate_all() { |
duke@435 | 3155 | // Generates all stubs and initializes the entry points |
duke@435 | 3156 | |
duke@435 | 3157 | // These entry points require SharedInfo::stack0 to be set up in non-core builds |
duke@435 | 3158 | // and need to be relocatable, so they each fabricate a RuntimeStub internally. |
never@3136 | 3159 | StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError)); |
never@3136 | 3160 | StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError)); |
never@3136 | 3161 | StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call)); |
duke@435 | 3162 | |
duke@435 | 3163 | //------------------------------------------------------------------------------------------------------------------------ |
duke@435 | 3164 | // entry points that are platform specific |
duke@435 | 3165 | |
duke@435 | 3166 | // support for verify_oop (must happen after universe_init) |
duke@435 | 3167 | StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop(); |
duke@435 | 3168 | |
duke@435 | 3169 | // arraycopy stubs used by compilers |
duke@435 | 3170 | generate_arraycopy_stubs(); |
jrose@1145 | 3171 | |
never@1609 | 3172 | generate_math_stubs(); |
kvn@4205 | 3173 | |
kvn@4205 | 3174 | // don't bother generating these AES intrinsic stubs unless global flag is set |
kvn@4205 | 3175 | if (UseAESIntrinsics) { |
kvn@4205 | 3176 | StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others |
kvn@4205 | 3177 | |
kvn@4205 | 3178 | StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock(); |
kvn@4205 | 3179 | StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock(); |
kvn@4205 | 3180 | StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt(); |
kvn@4205 | 3181 | StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt(); |
kvn@4205 | 3182 | } |
goetz@5400 | 3183 | |
ascarpino@9788 | 3184 | // Generate GHASH intrinsics code |
ascarpino@9788 | 3185 | if (UseGHASHIntrinsics) { |
ascarpino@9788 | 3186 | StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask(); |
ascarpino@9788 | 3187 | StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask(); |
ascarpino@9788 | 3188 | StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks(); |
ascarpino@9788 | 3189 | } |
ascarpino@9788 | 3190 | |
goetz@5400 | 3191 | // Safefetch stubs. |
goetz@5400 | 3192 | generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry, |
goetz@5400 | 3193 | &StubRoutines::_safefetch32_fault_pc, |
goetz@5400 | 3194 | &StubRoutines::_safefetch32_continuation_pc); |
goetz@5400 | 3195 | StubRoutines::_safefetchN_entry = StubRoutines::_safefetch32_entry; |
goetz@5400 | 3196 | StubRoutines::_safefetchN_fault_pc = StubRoutines::_safefetch32_fault_pc; |
goetz@5400 | 3197 | StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc; |
duke@435 | 3198 | } |
duke@435 | 3199 | |
duke@435 | 3200 | |
duke@435 | 3201 | public: |
duke@435 | 3202 | StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { |
duke@435 | 3203 | if (all) { |
duke@435 | 3204 | generate_all(); |
duke@435 | 3205 | } else { |
duke@435 | 3206 | generate_initial(); |
duke@435 | 3207 | } |
duke@435 | 3208 | } |
duke@435 | 3209 | }; // end class declaration |
duke@435 | 3210 | |
duke@435 | 3211 | |
duke@435 | 3212 | void StubGenerator_generate(CodeBuffer* code, bool all) { |
duke@435 | 3213 | StubGenerator g(code, all); |
duke@435 | 3214 | } |