Tue, 06 Dec 2011 18:28:51 -0500
7117052: instanceKlass::_init_state can be u1 type
Summary: Change instanceKlass::_init_state field to u1 type.
Reviewed-by: bdelsart, coleenp, dholmes, phh, never
Contributed-by: Jiangli Zhou <jiangli.zhou@oracle.com>
duke@435 | 1 | /* |
iveresov@2595 | 2 | * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved. |
duke@435 | 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
duke@435 | 4 | * |
duke@435 | 5 | * This code is free software; you can redistribute it and/or modify it |
duke@435 | 6 | * under the terms of the GNU General Public License version 2 only, as |
duke@435 | 7 | * published by the Free Software Foundation. |
duke@435 | 8 | * |
duke@435 | 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
duke@435 | 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
duke@435 | 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
duke@435 | 12 | * version 2 for more details (a copy is included in the LICENSE file that |
duke@435 | 13 | * accompanied this code). |
duke@435 | 14 | * |
duke@435 | 15 | * You should have received a copy of the GNU General Public License version |
duke@435 | 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
duke@435 | 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
duke@435 | 18 | * |
trims@1907 | 19 | * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
trims@1907 | 20 | * or visit www.oracle.com if you need additional information or have any |
trims@1907 | 21 | * questions. |
duke@435 | 22 | * |
duke@435 | 23 | */ |
duke@435 | 24 | |
stefank@2314 | 25 | #include "precompiled.hpp" |
stefank@2314 | 26 | #include "asm/assembler.hpp" |
stefank@2314 | 27 | #include "assembler_sparc.inline.hpp" |
stefank@2314 | 28 | #include "interpreter/interpreter.hpp" |
stefank@2314 | 29 | #include "nativeInst_sparc.hpp" |
stefank@2314 | 30 | #include "oops/instanceOop.hpp" |
stefank@2314 | 31 | #include "oops/methodOop.hpp" |
stefank@2314 | 32 | #include "oops/objArrayKlass.hpp" |
stefank@2314 | 33 | #include "oops/oop.inline.hpp" |
stefank@2314 | 34 | #include "prims/methodHandles.hpp" |
stefank@2314 | 35 | #include "runtime/frame.inline.hpp" |
stefank@2314 | 36 | #include "runtime/handles.inline.hpp" |
stefank@2314 | 37 | #include "runtime/sharedRuntime.hpp" |
stefank@2314 | 38 | #include "runtime/stubCodeGenerator.hpp" |
stefank@2314 | 39 | #include "runtime/stubRoutines.hpp" |
stefank@2314 | 40 | #include "utilities/top.hpp" |
stefank@2314 | 41 | #ifdef TARGET_OS_FAMILY_linux |
stefank@2314 | 42 | # include "thread_linux.inline.hpp" |
stefank@2314 | 43 | #endif |
stefank@2314 | 44 | #ifdef TARGET_OS_FAMILY_solaris |
stefank@2314 | 45 | # include "thread_solaris.inline.hpp" |
stefank@2314 | 46 | #endif |
stefank@2314 | 47 | #ifdef COMPILER2 |
stefank@2314 | 48 | #include "opto/runtime.hpp" |
stefank@2314 | 49 | #endif |
duke@435 | 50 | |
duke@435 | 51 | // Declaration and definition of StubGenerator (no .hpp file). |
duke@435 | 52 | // For a more detailed description of the stub routine structure |
duke@435 | 53 | // see the comment in stubRoutines.hpp. |
duke@435 | 54 | |
duke@435 | 55 | #define __ _masm-> |
duke@435 | 56 | |
duke@435 | 57 | #ifdef PRODUCT |
duke@435 | 58 | #define BLOCK_COMMENT(str) /* nothing */ |
duke@435 | 59 | #else |
duke@435 | 60 | #define BLOCK_COMMENT(str) __ block_comment(str) |
duke@435 | 61 | #endif |
duke@435 | 62 | |
duke@435 | 63 | #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") |
duke@435 | 64 | |
duke@435 | 65 | // Note: The register L7 is used as L7_thread_cache, and may not be used |
duke@435 | 66 | // any other way within this module. |
duke@435 | 67 | |
duke@435 | 68 | |
duke@435 | 69 | static const Register& Lstub_temp = L2; |
duke@435 | 70 | |
duke@435 | 71 | // ------------------------------------------------------------------------------------------------------------------------- |
duke@435 | 72 | // Stub Code definitions |
duke@435 | 73 | |
duke@435 | 74 | static address handle_unsafe_access() { |
duke@435 | 75 | JavaThread* thread = JavaThread::current(); |
duke@435 | 76 | address pc = thread->saved_exception_pc(); |
duke@435 | 77 | address npc = thread->saved_exception_npc(); |
duke@435 | 78 | // pc is the instruction which we must emulate |
duke@435 | 79 | // doing a no-op is fine: return garbage from the load |
duke@435 | 80 | |
duke@435 | 81 | // request an async exception |
duke@435 | 82 | thread->set_pending_unsafe_access_error(); |
duke@435 | 83 | |
duke@435 | 84 | // return address of next instruction to execute |
duke@435 | 85 | return npc; |
duke@435 | 86 | } |
duke@435 | 87 | |
duke@435 | 88 | class StubGenerator: public StubCodeGenerator { |
duke@435 | 89 | private: |
duke@435 | 90 | |
duke@435 | 91 | #ifdef PRODUCT |
duke@435 | 92 | #define inc_counter_np(a,b,c) (0) |
duke@435 | 93 | #else |
duke@435 | 94 | #define inc_counter_np(counter, t1, t2) \ |
duke@435 | 95 | BLOCK_COMMENT("inc_counter " #counter); \ |
twisti@1162 | 96 | __ inc_counter(&counter, t1, t2); |
duke@435 | 97 | #endif |
duke@435 | 98 | |
duke@435 | 99 | //---------------------------------------------------------------------------------------------------- |
duke@435 | 100 | // Call stubs are used to call Java from C |
duke@435 | 101 | |
duke@435 | 102 | address generate_call_stub(address& return_pc) { |
duke@435 | 103 | StubCodeMark mark(this, "StubRoutines", "call_stub"); |
duke@435 | 104 | address start = __ pc(); |
duke@435 | 105 | |
duke@435 | 106 | // Incoming arguments: |
duke@435 | 107 | // |
duke@435 | 108 | // o0 : call wrapper address |
duke@435 | 109 | // o1 : result (address) |
duke@435 | 110 | // o2 : result type |
duke@435 | 111 | // o3 : method |
duke@435 | 112 | // o4 : (interpreter) entry point |
duke@435 | 113 | // o5 : parameters (address) |
duke@435 | 114 | // [sp + 0x5c]: parameter size (in words) |
duke@435 | 115 | // [sp + 0x60]: thread |
duke@435 | 116 | // |
duke@435 | 117 | // +---------------+ <--- sp + 0 |
duke@435 | 118 | // | | |
duke@435 | 119 | // . reg save area . |
duke@435 | 120 | // | | |
duke@435 | 121 | // +---------------+ <--- sp + 0x40 |
duke@435 | 122 | // | | |
duke@435 | 123 | // . extra 7 slots . |
duke@435 | 124 | // | | |
duke@435 | 125 | // +---------------+ <--- sp + 0x5c |
duke@435 | 126 | // | param. size | |
duke@435 | 127 | // +---------------+ <--- sp + 0x60 |
duke@435 | 128 | // | thread | |
duke@435 | 129 | // +---------------+ |
duke@435 | 130 | // | | |
duke@435 | 131 | |
duke@435 | 132 | // note: if the link argument position changes, adjust |
duke@435 | 133 | // the code in frame::entry_frame_call_wrapper() |
duke@435 | 134 | |
duke@435 | 135 | const Argument link = Argument(0, false); // used only for GC |
duke@435 | 136 | const Argument result = Argument(1, false); |
duke@435 | 137 | const Argument result_type = Argument(2, false); |
duke@435 | 138 | const Argument method = Argument(3, false); |
duke@435 | 139 | const Argument entry_point = Argument(4, false); |
duke@435 | 140 | const Argument parameters = Argument(5, false); |
duke@435 | 141 | const Argument parameter_size = Argument(6, false); |
duke@435 | 142 | const Argument thread = Argument(7, false); |
duke@435 | 143 | |
duke@435 | 144 | // setup thread register |
duke@435 | 145 | __ ld_ptr(thread.as_address(), G2_thread); |
coleenp@548 | 146 | __ reinit_heapbase(); |
duke@435 | 147 | |
duke@435 | 148 | #ifdef ASSERT |
duke@435 | 149 | // make sure we have no pending exceptions |
duke@435 | 150 | { const Register t = G3_scratch; |
duke@435 | 151 | Label L; |
duke@435 | 152 | __ ld_ptr(G2_thread, in_bytes(Thread::pending_exception_offset()), t); |
kvn@3037 | 153 | __ br_null_short(t, Assembler::pt, L); |
duke@435 | 154 | __ stop("StubRoutines::call_stub: entered with pending exception"); |
duke@435 | 155 | __ bind(L); |
duke@435 | 156 | } |
duke@435 | 157 | #endif |
duke@435 | 158 | |
duke@435 | 159 | // create activation frame & allocate space for parameters |
duke@435 | 160 | { const Register t = G3_scratch; |
duke@435 | 161 | __ ld_ptr(parameter_size.as_address(), t); // get parameter size (in words) |
duke@435 | 162 | __ add(t, frame::memory_parameter_word_sp_offset, t); // add space for save area (in words) |
duke@435 | 163 | __ round_to(t, WordsPerLong); // make sure it is multiple of 2 (in words) |
twisti@1861 | 164 | __ sll(t, Interpreter::logStackElementSize, t); // compute number of bytes |
duke@435 | 165 | __ neg(t); // negate so it can be used with save |
duke@435 | 166 | __ save(SP, t, SP); // setup new frame |
duke@435 | 167 | } |
duke@435 | 168 | |
duke@435 | 169 | // +---------------+ <--- sp + 0 |
duke@435 | 170 | // | | |
duke@435 | 171 | // . reg save area . |
duke@435 | 172 | // | | |
duke@435 | 173 | // +---------------+ <--- sp + 0x40 |
duke@435 | 174 | // | | |
duke@435 | 175 | // . extra 7 slots . |
duke@435 | 176 | // | | |
duke@435 | 177 | // +---------------+ <--- sp + 0x5c |
duke@435 | 178 | // | empty slot | (only if parameter size is even) |
duke@435 | 179 | // +---------------+ |
duke@435 | 180 | // | | |
duke@435 | 181 | // . parameters . |
duke@435 | 182 | // | | |
duke@435 | 183 | // +---------------+ <--- fp + 0 |
duke@435 | 184 | // | | |
duke@435 | 185 | // . reg save area . |
duke@435 | 186 | // | | |
duke@435 | 187 | // +---------------+ <--- fp + 0x40 |
duke@435 | 188 | // | | |
duke@435 | 189 | // . extra 7 slots . |
duke@435 | 190 | // | | |
duke@435 | 191 | // +---------------+ <--- fp + 0x5c |
duke@435 | 192 | // | param. size | |
duke@435 | 193 | // +---------------+ <--- fp + 0x60 |
duke@435 | 194 | // | thread | |
duke@435 | 195 | // +---------------+ |
duke@435 | 196 | // | | |
duke@435 | 197 | |
duke@435 | 198 | // pass parameters if any |
duke@435 | 199 | BLOCK_COMMENT("pass parameters if any"); |
duke@435 | 200 | { const Register src = parameters.as_in().as_register(); |
duke@435 | 201 | const Register dst = Lentry_args; |
duke@435 | 202 | const Register tmp = G3_scratch; |
duke@435 | 203 | const Register cnt = G4_scratch; |
duke@435 | 204 | |
duke@435 | 205 | // test if any parameters & setup of Lentry_args |
duke@435 | 206 | Label exit; |
duke@435 | 207 | __ ld_ptr(parameter_size.as_in().as_address(), cnt); // parameter counter |
duke@435 | 208 | __ add( FP, STACK_BIAS, dst ); |
kvn@3037 | 209 | __ cmp_zero_and_br(Assembler::zero, cnt, exit); |
duke@435 | 210 | __ delayed()->sub(dst, BytesPerWord, dst); // setup Lentry_args |
duke@435 | 211 | |
duke@435 | 212 | // copy parameters if any |
duke@435 | 213 | Label loop; |
duke@435 | 214 | __ BIND(loop); |
duke@435 | 215 | // Store parameter value |
duke@435 | 216 | __ ld_ptr(src, 0, tmp); |
duke@435 | 217 | __ add(src, BytesPerWord, src); |
twisti@1861 | 218 | __ st_ptr(tmp, dst, 0); |
duke@435 | 219 | __ deccc(cnt); |
duke@435 | 220 | __ br(Assembler::greater, false, Assembler::pt, loop); |
twisti@1861 | 221 | __ delayed()->sub(dst, Interpreter::stackElementSize, dst); |
duke@435 | 222 | |
duke@435 | 223 | // done |
duke@435 | 224 | __ BIND(exit); |
duke@435 | 225 | } |
duke@435 | 226 | |
duke@435 | 227 | // setup parameters, method & call Java function |
duke@435 | 228 | #ifdef ASSERT |
duke@435 | 229 | // layout_activation_impl checks it's notion of saved SP against |
duke@435 | 230 | // this register, so if this changes update it as well. |
duke@435 | 231 | const Register saved_SP = Lscratch; |
duke@435 | 232 | __ mov(SP, saved_SP); // keep track of SP before call |
duke@435 | 233 | #endif |
duke@435 | 234 | |
duke@435 | 235 | // setup parameters |
duke@435 | 236 | const Register t = G3_scratch; |
duke@435 | 237 | __ ld_ptr(parameter_size.as_in().as_address(), t); // get parameter size (in words) |
twisti@1861 | 238 | __ sll(t, Interpreter::logStackElementSize, t); // compute number of bytes |
duke@435 | 239 | __ sub(FP, t, Gargs); // setup parameter pointer |
duke@435 | 240 | #ifdef _LP64 |
duke@435 | 241 | __ add( Gargs, STACK_BIAS, Gargs ); // Account for LP64 stack bias |
duke@435 | 242 | #endif |
duke@435 | 243 | __ mov(SP, O5_savedSP); |
duke@435 | 244 | |
duke@435 | 245 | |
duke@435 | 246 | // do the call |
duke@435 | 247 | // |
duke@435 | 248 | // the following register must be setup: |
duke@435 | 249 | // |
duke@435 | 250 | // G2_thread |
duke@435 | 251 | // G5_method |
duke@435 | 252 | // Gargs |
duke@435 | 253 | BLOCK_COMMENT("call Java function"); |
duke@435 | 254 | __ jmpl(entry_point.as_in().as_register(), G0, O7); |
duke@435 | 255 | __ delayed()->mov(method.as_in().as_register(), G5_method); // setup method |
duke@435 | 256 | |
duke@435 | 257 | BLOCK_COMMENT("call_stub_return_address:"); |
duke@435 | 258 | return_pc = __ pc(); |
duke@435 | 259 | |
duke@435 | 260 | // The callee, if it wasn't interpreted, can return with SP changed so |
duke@435 | 261 | // we can no longer assert of change of SP. |
duke@435 | 262 | |
duke@435 | 263 | // store result depending on type |
duke@435 | 264 | // (everything that is not T_OBJECT, T_LONG, T_FLOAT, or T_DOUBLE |
duke@435 | 265 | // is treated as T_INT) |
duke@435 | 266 | { const Register addr = result .as_in().as_register(); |
duke@435 | 267 | const Register type = result_type.as_in().as_register(); |
duke@435 | 268 | Label is_long, is_float, is_double, is_object, exit; |
duke@435 | 269 | __ cmp(type, T_OBJECT); __ br(Assembler::equal, false, Assembler::pn, is_object); |
duke@435 | 270 | __ delayed()->cmp(type, T_FLOAT); __ br(Assembler::equal, false, Assembler::pn, is_float); |
duke@435 | 271 | __ delayed()->cmp(type, T_DOUBLE); __ br(Assembler::equal, false, Assembler::pn, is_double); |
duke@435 | 272 | __ delayed()->cmp(type, T_LONG); __ br(Assembler::equal, false, Assembler::pn, is_long); |
duke@435 | 273 | __ delayed()->nop(); |
duke@435 | 274 | |
duke@435 | 275 | // store int result |
duke@435 | 276 | __ st(O0, addr, G0); |
duke@435 | 277 | |
duke@435 | 278 | __ BIND(exit); |
duke@435 | 279 | __ ret(); |
duke@435 | 280 | __ delayed()->restore(); |
duke@435 | 281 | |
duke@435 | 282 | __ BIND(is_object); |
kvn@3037 | 283 | __ ba(exit); |
duke@435 | 284 | __ delayed()->st_ptr(O0, addr, G0); |
duke@435 | 285 | |
duke@435 | 286 | __ BIND(is_float); |
kvn@3037 | 287 | __ ba(exit); |
duke@435 | 288 | __ delayed()->stf(FloatRegisterImpl::S, F0, addr, G0); |
duke@435 | 289 | |
duke@435 | 290 | __ BIND(is_double); |
kvn@3037 | 291 | __ ba(exit); |
duke@435 | 292 | __ delayed()->stf(FloatRegisterImpl::D, F0, addr, G0); |
duke@435 | 293 | |
duke@435 | 294 | __ BIND(is_long); |
duke@435 | 295 | #ifdef _LP64 |
kvn@3037 | 296 | __ ba(exit); |
duke@435 | 297 | __ delayed()->st_long(O0, addr, G0); // store entire long |
duke@435 | 298 | #else |
duke@435 | 299 | #if defined(COMPILER2) |
duke@435 | 300 | // All return values are where we want them, except for Longs. C2 returns |
duke@435 | 301 | // longs in G1 in the 32-bit build whereas the interpreter wants them in O0/O1. |
duke@435 | 302 | // Since the interpreter will return longs in G1 and O0/O1 in the 32bit |
duke@435 | 303 | // build we simply always use G1. |
duke@435 | 304 | // Note: I tried to make c2 return longs in O0/O1 and G1 so we wouldn't have to |
duke@435 | 305 | // do this here. Unfortunately if we did a rethrow we'd see an machepilog node |
duke@435 | 306 | // first which would move g1 -> O0/O1 and destroy the exception we were throwing. |
duke@435 | 307 | |
kvn@3037 | 308 | __ ba(exit); |
duke@435 | 309 | __ delayed()->stx(G1, addr, G0); // store entire long |
duke@435 | 310 | #else |
duke@435 | 311 | __ st(O1, addr, BytesPerInt); |
kvn@3037 | 312 | __ ba(exit); |
duke@435 | 313 | __ delayed()->st(O0, addr, G0); |
duke@435 | 314 | #endif /* COMPILER2 */ |
duke@435 | 315 | #endif /* _LP64 */ |
duke@435 | 316 | } |
duke@435 | 317 | return start; |
duke@435 | 318 | } |
duke@435 | 319 | |
duke@435 | 320 | |
duke@435 | 321 | //---------------------------------------------------------------------------------------------------- |
duke@435 | 322 | // Return point for a Java call if there's an exception thrown in Java code. |
duke@435 | 323 | // The exception is caught and transformed into a pending exception stored in |
duke@435 | 324 | // JavaThread that can be tested from within the VM. |
duke@435 | 325 | // |
duke@435 | 326 | // Oexception: exception oop |
duke@435 | 327 | |
duke@435 | 328 | address generate_catch_exception() { |
duke@435 | 329 | StubCodeMark mark(this, "StubRoutines", "catch_exception"); |
duke@435 | 330 | |
duke@435 | 331 | address start = __ pc(); |
duke@435 | 332 | // verify that thread corresponds |
duke@435 | 333 | __ verify_thread(); |
duke@435 | 334 | |
duke@435 | 335 | const Register& temp_reg = Gtemp; |
twisti@1162 | 336 | Address pending_exception_addr (G2_thread, Thread::pending_exception_offset()); |
twisti@1162 | 337 | Address exception_file_offset_addr(G2_thread, Thread::exception_file_offset ()); |
twisti@1162 | 338 | Address exception_line_offset_addr(G2_thread, Thread::exception_line_offset ()); |
duke@435 | 339 | |
duke@435 | 340 | // set pending exception |
duke@435 | 341 | __ verify_oop(Oexception); |
duke@435 | 342 | __ st_ptr(Oexception, pending_exception_addr); |
duke@435 | 343 | __ set((intptr_t)__FILE__, temp_reg); |
duke@435 | 344 | __ st_ptr(temp_reg, exception_file_offset_addr); |
duke@435 | 345 | __ set((intptr_t)__LINE__, temp_reg); |
duke@435 | 346 | __ st(temp_reg, exception_line_offset_addr); |
duke@435 | 347 | |
duke@435 | 348 | // complete return to VM |
duke@435 | 349 | assert(StubRoutines::_call_stub_return_address != NULL, "must have been generated before"); |
duke@435 | 350 | |
twisti@1162 | 351 | AddressLiteral stub_ret(StubRoutines::_call_stub_return_address); |
twisti@1162 | 352 | __ jump_to(stub_ret, temp_reg); |
duke@435 | 353 | __ delayed()->nop(); |
duke@435 | 354 | |
duke@435 | 355 | return start; |
duke@435 | 356 | } |
duke@435 | 357 | |
duke@435 | 358 | |
duke@435 | 359 | //---------------------------------------------------------------------------------------------------- |
duke@435 | 360 | // Continuation point for runtime calls returning with a pending exception |
duke@435 | 361 | // The pending exception check happened in the runtime or native call stub |
duke@435 | 362 | // The pending exception in Thread is converted into a Java-level exception |
duke@435 | 363 | // |
duke@435 | 364 | // Contract with Java-level exception handler: O0 = exception |
duke@435 | 365 | // O1 = throwing pc |
duke@435 | 366 | |
duke@435 | 367 | address generate_forward_exception() { |
duke@435 | 368 | StubCodeMark mark(this, "StubRoutines", "forward_exception"); |
duke@435 | 369 | address start = __ pc(); |
duke@435 | 370 | |
duke@435 | 371 | // Upon entry, O7 has the return address returning into Java |
duke@435 | 372 | // (interpreted or compiled) code; i.e. the return address |
duke@435 | 373 | // becomes the throwing pc. |
duke@435 | 374 | |
duke@435 | 375 | const Register& handler_reg = Gtemp; |
duke@435 | 376 | |
twisti@1162 | 377 | Address exception_addr(G2_thread, Thread::pending_exception_offset()); |
duke@435 | 378 | |
duke@435 | 379 | #ifdef ASSERT |
duke@435 | 380 | // make sure that this code is only executed if there is a pending exception |
duke@435 | 381 | { Label L; |
duke@435 | 382 | __ ld_ptr(exception_addr, Gtemp); |
kvn@3037 | 383 | __ br_notnull_short(Gtemp, Assembler::pt, L); |
duke@435 | 384 | __ stop("StubRoutines::forward exception: no pending exception (1)"); |
duke@435 | 385 | __ bind(L); |
duke@435 | 386 | } |
duke@435 | 387 | #endif |
duke@435 | 388 | |
duke@435 | 389 | // compute exception handler into handler_reg |
duke@435 | 390 | __ get_thread(); |
duke@435 | 391 | __ ld_ptr(exception_addr, Oexception); |
duke@435 | 392 | __ verify_oop(Oexception); |
duke@435 | 393 | __ save_frame(0); // compensates for compiler weakness |
duke@435 | 394 | __ add(O7->after_save(), frame::pc_return_offset, Lscratch); // save the issuing PC |
duke@435 | 395 | BLOCK_COMMENT("call exception_handler_for_return_address"); |
twisti@1730 | 396 | __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), G2_thread, Lscratch); |
duke@435 | 397 | __ mov(O0, handler_reg); |
duke@435 | 398 | __ restore(); // compensates for compiler weakness |
duke@435 | 399 | |
duke@435 | 400 | __ ld_ptr(exception_addr, Oexception); |
duke@435 | 401 | __ add(O7, frame::pc_return_offset, Oissuing_pc); // save the issuing PC |
duke@435 | 402 | |
duke@435 | 403 | #ifdef ASSERT |
duke@435 | 404 | // make sure exception is set |
duke@435 | 405 | { Label L; |
kvn@3037 | 406 | __ br_notnull_short(Oexception, Assembler::pt, L); |
duke@435 | 407 | __ stop("StubRoutines::forward exception: no pending exception (2)"); |
duke@435 | 408 | __ bind(L); |
duke@435 | 409 | } |
duke@435 | 410 | #endif |
duke@435 | 411 | // jump to exception handler |
duke@435 | 412 | __ jmp(handler_reg, 0); |
duke@435 | 413 | // clear pending exception |
duke@435 | 414 | __ delayed()->st_ptr(G0, exception_addr); |
duke@435 | 415 | |
duke@435 | 416 | return start; |
duke@435 | 417 | } |
duke@435 | 418 | |
duke@435 | 419 | |
duke@435 | 420 | //------------------------------------------------------------------------------------------------------------------------ |
duke@435 | 421 | // Continuation point for throwing of implicit exceptions that are not handled in |
duke@435 | 422 | // the current activation. Fabricates an exception oop and initiates normal |
duke@435 | 423 | // exception dispatching in this frame. Only callee-saved registers are preserved |
duke@435 | 424 | // (through the normal register window / RegisterMap handling). |
duke@435 | 425 | // If the compiler needs all registers to be preserved between the fault |
duke@435 | 426 | // point and the exception handler then it must assume responsibility for that in |
duke@435 | 427 | // AbstractCompiler::continuation_for_implicit_null_exception or |
duke@435 | 428 | // continuation_for_implicit_division_by_zero_exception. All other implicit |
duke@435 | 429 | // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are |
duke@435 | 430 | // either at call sites or otherwise assume that stack unwinding will be initiated, |
duke@435 | 431 | // so caller saved registers were assumed volatile in the compiler. |
duke@435 | 432 | |
duke@435 | 433 | // Note that we generate only this stub into a RuntimeStub, because it needs to be |
duke@435 | 434 | // properly traversed and ignored during GC, so we change the meaning of the "__" |
duke@435 | 435 | // macro within this method. |
duke@435 | 436 | #undef __ |
duke@435 | 437 | #define __ masm-> |
duke@435 | 438 | |
never@3136 | 439 | address generate_throw_exception(const char* name, address runtime_entry, |
never@2978 | 440 | Register arg1 = noreg, Register arg2 = noreg) { |
duke@435 | 441 | #ifdef ASSERT |
duke@435 | 442 | int insts_size = VerifyThread ? 1 * K : 600; |
duke@435 | 443 | #else |
duke@435 | 444 | int insts_size = VerifyThread ? 1 * K : 256; |
duke@435 | 445 | #endif /* ASSERT */ |
duke@435 | 446 | int locs_size = 32; |
duke@435 | 447 | |
duke@435 | 448 | CodeBuffer code(name, insts_size, locs_size); |
duke@435 | 449 | MacroAssembler* masm = new MacroAssembler(&code); |
duke@435 | 450 | |
duke@435 | 451 | __ verify_thread(); |
duke@435 | 452 | |
duke@435 | 453 | // This is an inlined and slightly modified version of call_VM |
duke@435 | 454 | // which has the ability to fetch the return PC out of thread-local storage |
duke@435 | 455 | __ assert_not_delayed(); |
duke@435 | 456 | |
duke@435 | 457 | // Note that we always push a frame because on the SPARC |
duke@435 | 458 | // architecture, for all of our implicit exception kinds at call |
duke@435 | 459 | // sites, the implicit exception is taken before the callee frame |
duke@435 | 460 | // is pushed. |
duke@435 | 461 | __ save_frame(0); |
duke@435 | 462 | |
duke@435 | 463 | int frame_complete = __ offset(); |
duke@435 | 464 | |
duke@435 | 465 | // Note that we always have a runtime stub frame on the top of stack by this point |
duke@435 | 466 | Register last_java_sp = SP; |
duke@435 | 467 | // 64-bit last_java_sp is biased! |
duke@435 | 468 | __ set_last_Java_frame(last_java_sp, G0); |
duke@435 | 469 | if (VerifyThread) __ mov(G2_thread, O0); // about to be smashed; pass early |
duke@435 | 470 | __ save_thread(noreg); |
never@2978 | 471 | if (arg1 != noreg) { |
never@2978 | 472 | assert(arg2 != O1, "clobbered"); |
never@2978 | 473 | __ mov(arg1, O1); |
never@2978 | 474 | } |
never@2978 | 475 | if (arg2 != noreg) { |
never@2978 | 476 | __ mov(arg2, O2); |
never@2978 | 477 | } |
duke@435 | 478 | // do the call |
duke@435 | 479 | BLOCK_COMMENT("call runtime_entry"); |
duke@435 | 480 | __ call(runtime_entry, relocInfo::runtime_call_type); |
duke@435 | 481 | if (!VerifyThread) |
duke@435 | 482 | __ delayed()->mov(G2_thread, O0); // pass thread as first argument |
duke@435 | 483 | else |
duke@435 | 484 | __ delayed()->nop(); // (thread already passed) |
duke@435 | 485 | __ restore_thread(noreg); |
duke@435 | 486 | __ reset_last_Java_frame(); |
duke@435 | 487 | |
duke@435 | 488 | // check for pending exceptions. use Gtemp as scratch register. |
duke@435 | 489 | #ifdef ASSERT |
duke@435 | 490 | Label L; |
duke@435 | 491 | |
twisti@1162 | 492 | Address exception_addr(G2_thread, Thread::pending_exception_offset()); |
duke@435 | 493 | Register scratch_reg = Gtemp; |
duke@435 | 494 | __ ld_ptr(exception_addr, scratch_reg); |
kvn@3037 | 495 | __ br_notnull_short(scratch_reg, Assembler::pt, L); |
duke@435 | 496 | __ should_not_reach_here(); |
duke@435 | 497 | __ bind(L); |
duke@435 | 498 | #endif // ASSERT |
duke@435 | 499 | BLOCK_COMMENT("call forward_exception_entry"); |
duke@435 | 500 | __ call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type); |
duke@435 | 501 | // we use O7 linkage so that forward_exception_entry has the issuing PC |
duke@435 | 502 | __ delayed()->restore(); |
duke@435 | 503 | |
duke@435 | 504 | RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, masm->total_frame_size_in_bytes(0), NULL, false); |
duke@435 | 505 | return stub->entry_point(); |
duke@435 | 506 | } |
duke@435 | 507 | |
duke@435 | 508 | #undef __ |
duke@435 | 509 | #define __ _masm-> |
duke@435 | 510 | |
duke@435 | 511 | |
duke@435 | 512 | // Generate a routine that sets all the registers so we |
duke@435 | 513 | // can tell if the stop routine prints them correctly. |
duke@435 | 514 | address generate_test_stop() { |
duke@435 | 515 | StubCodeMark mark(this, "StubRoutines", "test_stop"); |
duke@435 | 516 | address start = __ pc(); |
duke@435 | 517 | |
duke@435 | 518 | int i; |
duke@435 | 519 | |
duke@435 | 520 | __ save_frame(0); |
duke@435 | 521 | |
duke@435 | 522 | static jfloat zero = 0.0, one = 1.0; |
duke@435 | 523 | |
duke@435 | 524 | // put addr in L0, then load through L0 to F0 |
duke@435 | 525 | __ set((intptr_t)&zero, L0); __ ldf( FloatRegisterImpl::S, L0, 0, F0); |
duke@435 | 526 | __ set((intptr_t)&one, L0); __ ldf( FloatRegisterImpl::S, L0, 0, F1); // 1.0 to F1 |
duke@435 | 527 | |
duke@435 | 528 | // use add to put 2..18 in F2..F18 |
duke@435 | 529 | for ( i = 2; i <= 18; ++i ) { |
duke@435 | 530 | __ fadd( FloatRegisterImpl::S, F1, as_FloatRegister(i-1), as_FloatRegister(i)); |
duke@435 | 531 | } |
duke@435 | 532 | |
duke@435 | 533 | // Now put double 2 in F16, double 18 in F18 |
duke@435 | 534 | __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, F2, F16 ); |
duke@435 | 535 | __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, F18, F18 ); |
duke@435 | 536 | |
duke@435 | 537 | // use add to put 20..32 in F20..F32 |
duke@435 | 538 | for (i = 20; i < 32; i += 2) { |
duke@435 | 539 | __ fadd( FloatRegisterImpl::D, F16, as_FloatRegister(i-2), as_FloatRegister(i)); |
duke@435 | 540 | } |
duke@435 | 541 | |
duke@435 | 542 | // put 0..7 in i's, 8..15 in l's, 16..23 in o's, 24..31 in g's |
duke@435 | 543 | for ( i = 0; i < 8; ++i ) { |
duke@435 | 544 | if (i < 6) { |
duke@435 | 545 | __ set( i, as_iRegister(i)); |
duke@435 | 546 | __ set(16 + i, as_oRegister(i)); |
duke@435 | 547 | __ set(24 + i, as_gRegister(i)); |
duke@435 | 548 | } |
duke@435 | 549 | __ set( 8 + i, as_lRegister(i)); |
duke@435 | 550 | } |
duke@435 | 551 | |
duke@435 | 552 | __ stop("testing stop"); |
duke@435 | 553 | |
duke@435 | 554 | |
duke@435 | 555 | __ ret(); |
duke@435 | 556 | __ delayed()->restore(); |
duke@435 | 557 | |
duke@435 | 558 | return start; |
duke@435 | 559 | } |
duke@435 | 560 | |
duke@435 | 561 | |
duke@435 | 562 | address generate_stop_subroutine() { |
duke@435 | 563 | StubCodeMark mark(this, "StubRoutines", "stop_subroutine"); |
duke@435 | 564 | address start = __ pc(); |
duke@435 | 565 | |
duke@435 | 566 | __ stop_subroutine(); |
duke@435 | 567 | |
duke@435 | 568 | return start; |
duke@435 | 569 | } |
duke@435 | 570 | |
duke@435 | 571 | address generate_flush_callers_register_windows() { |
duke@435 | 572 | StubCodeMark mark(this, "StubRoutines", "flush_callers_register_windows"); |
duke@435 | 573 | address start = __ pc(); |
duke@435 | 574 | |
duke@435 | 575 | __ flush_windows(); |
duke@435 | 576 | __ retl(false); |
duke@435 | 577 | __ delayed()->add( FP, STACK_BIAS, O0 ); |
duke@435 | 578 | // The returned value must be a stack pointer whose register save area |
duke@435 | 579 | // is flushed, and will stay flushed while the caller executes. |
duke@435 | 580 | |
duke@435 | 581 | return start; |
duke@435 | 582 | } |
duke@435 | 583 | |
duke@435 | 584 | // Helper functions for v8 atomic operations. |
duke@435 | 585 | // |
duke@435 | 586 | void get_v8_oop_lock_ptr(Register lock_ptr_reg, Register mark_oop_reg, Register scratch_reg) { |
duke@435 | 587 | if (mark_oop_reg == noreg) { |
duke@435 | 588 | address lock_ptr = (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr(); |
duke@435 | 589 | __ set((intptr_t)lock_ptr, lock_ptr_reg); |
duke@435 | 590 | } else { |
duke@435 | 591 | assert(scratch_reg != noreg, "just checking"); |
duke@435 | 592 | address lock_ptr = (address)StubRoutines::Sparc::_v8_oop_lock_cache; |
duke@435 | 593 | __ set((intptr_t)lock_ptr, lock_ptr_reg); |
duke@435 | 594 | __ and3(mark_oop_reg, StubRoutines::Sparc::v8_oop_lock_mask_in_place, scratch_reg); |
duke@435 | 595 | __ add(lock_ptr_reg, scratch_reg, lock_ptr_reg); |
duke@435 | 596 | } |
duke@435 | 597 | } |
duke@435 | 598 | |
duke@435 | 599 | void generate_v8_lock_prologue(Register lock_reg, Register lock_ptr_reg, Register yield_reg, Label& retry, Label& dontyield, Register mark_oop_reg = noreg, Register scratch_reg = noreg) { |
duke@435 | 600 | |
duke@435 | 601 | get_v8_oop_lock_ptr(lock_ptr_reg, mark_oop_reg, scratch_reg); |
duke@435 | 602 | __ set(StubRoutines::Sparc::locked, lock_reg); |
duke@435 | 603 | // Initialize yield counter |
duke@435 | 604 | __ mov(G0,yield_reg); |
duke@435 | 605 | |
duke@435 | 606 | __ BIND(retry); |
kvn@3037 | 607 | __ cmp_and_br_short(yield_reg, V8AtomicOperationUnderLockSpinCount, Assembler::less, Assembler::pt, dontyield); |
duke@435 | 608 | |
duke@435 | 609 | // This code can only be called from inside the VM, this |
duke@435 | 610 | // stub is only invoked from Atomic::add(). We do not |
duke@435 | 611 | // want to use call_VM, because _last_java_sp and such |
duke@435 | 612 | // must already be set. |
duke@435 | 613 | // |
duke@435 | 614 | // Save the regs and make space for a C call |
duke@435 | 615 | __ save(SP, -96, SP); |
duke@435 | 616 | __ save_all_globals_into_locals(); |
duke@435 | 617 | BLOCK_COMMENT("call os::naked_sleep"); |
duke@435 | 618 | __ call(CAST_FROM_FN_PTR(address, os::naked_sleep)); |
duke@435 | 619 | __ delayed()->nop(); |
duke@435 | 620 | __ restore_globals_from_locals(); |
duke@435 | 621 | __ restore(); |
duke@435 | 622 | // reset the counter |
duke@435 | 623 | __ mov(G0,yield_reg); |
duke@435 | 624 | |
duke@435 | 625 | __ BIND(dontyield); |
duke@435 | 626 | |
duke@435 | 627 | // try to get lock |
duke@435 | 628 | __ swap(lock_ptr_reg, 0, lock_reg); |
duke@435 | 629 | |
duke@435 | 630 | // did we get the lock? |
duke@435 | 631 | __ cmp(lock_reg, StubRoutines::Sparc::unlocked); |
duke@435 | 632 | __ br(Assembler::notEqual, true, Assembler::pn, retry); |
duke@435 | 633 | __ delayed()->add(yield_reg,1,yield_reg); |
duke@435 | 634 | |
duke@435 | 635 | // yes, got lock. do the operation here. |
duke@435 | 636 | } |
duke@435 | 637 | |
duke@435 | 638 | void generate_v8_lock_epilogue(Register lock_reg, Register lock_ptr_reg, Register yield_reg, Label& retry, Label& dontyield, Register mark_oop_reg = noreg, Register scratch_reg = noreg) { |
duke@435 | 639 | __ st(lock_reg, lock_ptr_reg, 0); // unlock |
duke@435 | 640 | } |
duke@435 | 641 | |
duke@435 | 642 | // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest). |
duke@435 | 643 | // |
duke@435 | 644 | // Arguments : |
duke@435 | 645 | // |
duke@435 | 646 | // exchange_value: O0 |
duke@435 | 647 | // dest: O1 |
duke@435 | 648 | // |
duke@435 | 649 | // Results: |
duke@435 | 650 | // |
duke@435 | 651 | // O0: the value previously stored in dest |
duke@435 | 652 | // |
duke@435 | 653 | address generate_atomic_xchg() { |
duke@435 | 654 | StubCodeMark mark(this, "StubRoutines", "atomic_xchg"); |
duke@435 | 655 | address start = __ pc(); |
duke@435 | 656 | |
duke@435 | 657 | if (UseCASForSwap) { |
duke@435 | 658 | // Use CAS instead of swap, just in case the MP hardware |
duke@435 | 659 | // prefers to work with just one kind of synch. instruction. |
duke@435 | 660 | Label retry; |
duke@435 | 661 | __ BIND(retry); |
duke@435 | 662 | __ mov(O0, O3); // scratch copy of exchange value |
duke@435 | 663 | __ ld(O1, 0, O2); // observe the previous value |
duke@435 | 664 | // try to replace O2 with O3 |
duke@435 | 665 | __ cas_under_lock(O1, O2, O3, |
duke@435 | 666 | (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr(),false); |
kvn@3037 | 667 | __ cmp_and_br_short(O2, O3, Assembler::notEqual, Assembler::pn, retry); |
duke@435 | 668 | |
duke@435 | 669 | __ retl(false); |
duke@435 | 670 | __ delayed()->mov(O2, O0); // report previous value to caller |
duke@435 | 671 | |
duke@435 | 672 | } else { |
duke@435 | 673 | if (VM_Version::v9_instructions_work()) { |
duke@435 | 674 | __ retl(false); |
duke@435 | 675 | __ delayed()->swap(O1, 0, O0); |
duke@435 | 676 | } else { |
duke@435 | 677 | const Register& lock_reg = O2; |
duke@435 | 678 | const Register& lock_ptr_reg = O3; |
duke@435 | 679 | const Register& yield_reg = O4; |
duke@435 | 680 | |
duke@435 | 681 | Label retry; |
duke@435 | 682 | Label dontyield; |
duke@435 | 683 | |
duke@435 | 684 | generate_v8_lock_prologue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield); |
duke@435 | 685 | // got the lock, do the swap |
duke@435 | 686 | __ swap(O1, 0, O0); |
duke@435 | 687 | |
duke@435 | 688 | generate_v8_lock_epilogue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield); |
duke@435 | 689 | __ retl(false); |
duke@435 | 690 | __ delayed()->nop(); |
duke@435 | 691 | } |
duke@435 | 692 | } |
duke@435 | 693 | |
duke@435 | 694 | return start; |
duke@435 | 695 | } |
duke@435 | 696 | |
duke@435 | 697 | |
duke@435 | 698 | // Support for jint Atomic::cmpxchg(jint exchange_value, volatile jint* dest, jint compare_value) |
duke@435 | 699 | // |
duke@435 | 700 | // Arguments : |
duke@435 | 701 | // |
duke@435 | 702 | // exchange_value: O0 |
duke@435 | 703 | // dest: O1 |
duke@435 | 704 | // compare_value: O2 |
duke@435 | 705 | // |
duke@435 | 706 | // Results: |
duke@435 | 707 | // |
duke@435 | 708 | // O0: the value previously stored in dest |
duke@435 | 709 | // |
duke@435 | 710 | // Overwrites (v8): O3,O4,O5 |
duke@435 | 711 | // |
duke@435 | 712 | address generate_atomic_cmpxchg() { |
duke@435 | 713 | StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg"); |
duke@435 | 714 | address start = __ pc(); |
duke@435 | 715 | |
duke@435 | 716 | // cmpxchg(dest, compare_value, exchange_value) |
duke@435 | 717 | __ cas_under_lock(O1, O2, O0, |
duke@435 | 718 | (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr(),false); |
duke@435 | 719 | __ retl(false); |
duke@435 | 720 | __ delayed()->nop(); |
duke@435 | 721 | |
duke@435 | 722 | return start; |
duke@435 | 723 | } |
duke@435 | 724 | |
duke@435 | 725 | // Support for jlong Atomic::cmpxchg(jlong exchange_value, volatile jlong *dest, jlong compare_value) |
duke@435 | 726 | // |
duke@435 | 727 | // Arguments : |
duke@435 | 728 | // |
duke@435 | 729 | // exchange_value: O1:O0 |
duke@435 | 730 | // dest: O2 |
duke@435 | 731 | // compare_value: O4:O3 |
duke@435 | 732 | // |
duke@435 | 733 | // Results: |
duke@435 | 734 | // |
duke@435 | 735 | // O1:O0: the value previously stored in dest |
duke@435 | 736 | // |
duke@435 | 737 | // This only works on V9, on V8 we don't generate any |
duke@435 | 738 | // code and just return NULL. |
duke@435 | 739 | // |
duke@435 | 740 | // Overwrites: G1,G2,G3 |
duke@435 | 741 | // |
duke@435 | 742 | address generate_atomic_cmpxchg_long() { |
duke@435 | 743 | StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long"); |
duke@435 | 744 | address start = __ pc(); |
duke@435 | 745 | |
duke@435 | 746 | if (!VM_Version::supports_cx8()) |
duke@435 | 747 | return NULL;; |
duke@435 | 748 | __ sllx(O0, 32, O0); |
duke@435 | 749 | __ srl(O1, 0, O1); |
duke@435 | 750 | __ or3(O0,O1,O0); // O0 holds 64-bit value from compare_value |
duke@435 | 751 | __ sllx(O3, 32, O3); |
duke@435 | 752 | __ srl(O4, 0, O4); |
duke@435 | 753 | __ or3(O3,O4,O3); // O3 holds 64-bit value from exchange_value |
duke@435 | 754 | __ casx(O2, O3, O0); |
duke@435 | 755 | __ srl(O0, 0, O1); // unpacked return value in O1:O0 |
duke@435 | 756 | __ retl(false); |
duke@435 | 757 | __ delayed()->srlx(O0, 32, O0); |
duke@435 | 758 | |
duke@435 | 759 | return start; |
duke@435 | 760 | } |
duke@435 | 761 | |
duke@435 | 762 | |
duke@435 | 763 | // Support for jint Atomic::add(jint add_value, volatile jint* dest). |
duke@435 | 764 | // |
duke@435 | 765 | // Arguments : |
duke@435 | 766 | // |
duke@435 | 767 | // add_value: O0 (e.g., +1 or -1) |
duke@435 | 768 | // dest: O1 |
duke@435 | 769 | // |
duke@435 | 770 | // Results: |
duke@435 | 771 | // |
duke@435 | 772 | // O0: the new value stored in dest |
duke@435 | 773 | // |
duke@435 | 774 | // Overwrites (v9): O3 |
duke@435 | 775 | // Overwrites (v8): O3,O4,O5 |
duke@435 | 776 | // |
duke@435 | 777 | address generate_atomic_add() { |
duke@435 | 778 | StubCodeMark mark(this, "StubRoutines", "atomic_add"); |
duke@435 | 779 | address start = __ pc(); |
duke@435 | 780 | __ BIND(_atomic_add_stub); |
duke@435 | 781 | |
duke@435 | 782 | if (VM_Version::v9_instructions_work()) { |
duke@435 | 783 | Label(retry); |
duke@435 | 784 | __ BIND(retry); |
duke@435 | 785 | |
duke@435 | 786 | __ lduw(O1, 0, O2); |
kvn@3037 | 787 | __ add(O0, O2, O3); |
kvn@3037 | 788 | __ cas(O1, O2, O3); |
kvn@3037 | 789 | __ cmp_and_br_short(O2, O3, Assembler::notEqual, Assembler::pn, retry); |
duke@435 | 790 | __ retl(false); |
duke@435 | 791 | __ delayed()->add(O0, O2, O0); // note that cas made O2==O3 |
duke@435 | 792 | } else { |
duke@435 | 793 | const Register& lock_reg = O2; |
duke@435 | 794 | const Register& lock_ptr_reg = O3; |
duke@435 | 795 | const Register& value_reg = O4; |
duke@435 | 796 | const Register& yield_reg = O5; |
duke@435 | 797 | |
duke@435 | 798 | Label(retry); |
duke@435 | 799 | Label(dontyield); |
duke@435 | 800 | |
duke@435 | 801 | generate_v8_lock_prologue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield); |
duke@435 | 802 | // got lock, do the increment |
duke@435 | 803 | __ ld(O1, 0, value_reg); |
duke@435 | 804 | __ add(O0, value_reg, value_reg); |
duke@435 | 805 | __ st(value_reg, O1, 0); |
duke@435 | 806 | |
duke@435 | 807 | // %%% only for RMO and PSO |
duke@435 | 808 | __ membar(Assembler::StoreStore); |
duke@435 | 809 | |
duke@435 | 810 | generate_v8_lock_epilogue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield); |
duke@435 | 811 | |
duke@435 | 812 | __ retl(false); |
duke@435 | 813 | __ delayed()->mov(value_reg, O0); |
duke@435 | 814 | } |
duke@435 | 815 | |
duke@435 | 816 | return start; |
duke@435 | 817 | } |
duke@435 | 818 | Label _atomic_add_stub; // called from other stubs |
duke@435 | 819 | |
duke@435 | 820 | |
duke@435 | 821 | //------------------------------------------------------------------------------------------------------------------------ |
duke@435 | 822 | // The following routine generates a subroutine to throw an asynchronous |
duke@435 | 823 | // UnknownError when an unsafe access gets a fault that could not be |
duke@435 | 824 | // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.) |
duke@435 | 825 | // |
duke@435 | 826 | // Arguments : |
duke@435 | 827 | // |
duke@435 | 828 | // trapping PC: O7 |
duke@435 | 829 | // |
duke@435 | 830 | // Results: |
duke@435 | 831 | // posts an asynchronous exception, skips the trapping instruction |
duke@435 | 832 | // |
duke@435 | 833 | |
duke@435 | 834 | address generate_handler_for_unsafe_access() { |
duke@435 | 835 | StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access"); |
duke@435 | 836 | address start = __ pc(); |
duke@435 | 837 | |
duke@435 | 838 | const int preserve_register_words = (64 * 2); |
twisti@1162 | 839 | Address preserve_addr(FP, (-preserve_register_words * wordSize) + STACK_BIAS); |
duke@435 | 840 | |
duke@435 | 841 | Register Lthread = L7_thread_cache; |
duke@435 | 842 | int i; |
duke@435 | 843 | |
duke@435 | 844 | __ save_frame(0); |
duke@435 | 845 | __ mov(G1, L1); |
duke@435 | 846 | __ mov(G2, L2); |
duke@435 | 847 | __ mov(G3, L3); |
duke@435 | 848 | __ mov(G4, L4); |
duke@435 | 849 | __ mov(G5, L5); |
duke@435 | 850 | for (i = 0; i < (VM_Version::v9_instructions_work() ? 64 : 32); i += 2) { |
duke@435 | 851 | __ stf(FloatRegisterImpl::D, as_FloatRegister(i), preserve_addr, i * wordSize); |
duke@435 | 852 | } |
duke@435 | 853 | |
duke@435 | 854 | address entry_point = CAST_FROM_FN_PTR(address, handle_unsafe_access); |
duke@435 | 855 | BLOCK_COMMENT("call handle_unsafe_access"); |
duke@435 | 856 | __ call(entry_point, relocInfo::runtime_call_type); |
duke@435 | 857 | __ delayed()->nop(); |
duke@435 | 858 | |
duke@435 | 859 | __ mov(L1, G1); |
duke@435 | 860 | __ mov(L2, G2); |
duke@435 | 861 | __ mov(L3, G3); |
duke@435 | 862 | __ mov(L4, G4); |
duke@435 | 863 | __ mov(L5, G5); |
duke@435 | 864 | for (i = 0; i < (VM_Version::v9_instructions_work() ? 64 : 32); i += 2) { |
duke@435 | 865 | __ ldf(FloatRegisterImpl::D, preserve_addr, as_FloatRegister(i), i * wordSize); |
duke@435 | 866 | } |
duke@435 | 867 | |
duke@435 | 868 | __ verify_thread(); |
duke@435 | 869 | |
duke@435 | 870 | __ jmp(O0, 0); |
duke@435 | 871 | __ delayed()->restore(); |
duke@435 | 872 | |
duke@435 | 873 | return start; |
duke@435 | 874 | } |
duke@435 | 875 | |
duke@435 | 876 | |
duke@435 | 877 | // Support for uint StubRoutine::Sparc::partial_subtype_check( Klass sub, Klass super ); |
duke@435 | 878 | // Arguments : |
duke@435 | 879 | // |
duke@435 | 880 | // ret : O0, returned |
duke@435 | 881 | // icc/xcc: set as O0 (depending on wordSize) |
duke@435 | 882 | // sub : O1, argument, not changed |
duke@435 | 883 | // super: O2, argument, not changed |
duke@435 | 884 | // raddr: O7, blown by call |
duke@435 | 885 | address generate_partial_subtype_check() { |
coleenp@548 | 886 | __ align(CodeEntryAlignment); |
duke@435 | 887 | StubCodeMark mark(this, "StubRoutines", "partial_subtype_check"); |
duke@435 | 888 | address start = __ pc(); |
jrose@1079 | 889 | Label miss; |
duke@435 | 890 | |
duke@435 | 891 | #if defined(COMPILER2) && !defined(_LP64) |
duke@435 | 892 | // Do not use a 'save' because it blows the 64-bit O registers. |
coleenp@548 | 893 | __ add(SP,-4*wordSize,SP); // Make space for 4 temps (stack must be 2 words aligned) |
duke@435 | 894 | __ st_ptr(L0,SP,(frame::register_save_words+0)*wordSize); |
duke@435 | 895 | __ st_ptr(L1,SP,(frame::register_save_words+1)*wordSize); |
duke@435 | 896 | __ st_ptr(L2,SP,(frame::register_save_words+2)*wordSize); |
duke@435 | 897 | __ st_ptr(L3,SP,(frame::register_save_words+3)*wordSize); |
duke@435 | 898 | Register Rret = O0; |
duke@435 | 899 | Register Rsub = O1; |
duke@435 | 900 | Register Rsuper = O2; |
duke@435 | 901 | #else |
duke@435 | 902 | __ save_frame(0); |
duke@435 | 903 | Register Rret = I0; |
duke@435 | 904 | Register Rsub = I1; |
duke@435 | 905 | Register Rsuper = I2; |
duke@435 | 906 | #endif |
duke@435 | 907 | |
duke@435 | 908 | Register L0_ary_len = L0; |
duke@435 | 909 | Register L1_ary_ptr = L1; |
duke@435 | 910 | Register L2_super = L2; |
duke@435 | 911 | Register L3_index = L3; |
duke@435 | 912 | |
jrose@1079 | 913 | __ check_klass_subtype_slow_path(Rsub, Rsuper, |
jrose@1079 | 914 | L0, L1, L2, L3, |
jrose@1079 | 915 | NULL, &miss); |
jrose@1079 | 916 | |
jrose@1079 | 917 | // Match falls through here. |
jrose@1079 | 918 | __ addcc(G0,0,Rret); // set Z flags, Z result |
duke@435 | 919 | |
duke@435 | 920 | #if defined(COMPILER2) && !defined(_LP64) |
duke@435 | 921 | __ ld_ptr(SP,(frame::register_save_words+0)*wordSize,L0); |
duke@435 | 922 | __ ld_ptr(SP,(frame::register_save_words+1)*wordSize,L1); |
duke@435 | 923 | __ ld_ptr(SP,(frame::register_save_words+2)*wordSize,L2); |
duke@435 | 924 | __ ld_ptr(SP,(frame::register_save_words+3)*wordSize,L3); |
duke@435 | 925 | __ retl(); // Result in Rret is zero; flags set to Z |
duke@435 | 926 | __ delayed()->add(SP,4*wordSize,SP); |
duke@435 | 927 | #else |
duke@435 | 928 | __ ret(); // Result in Rret is zero; flags set to Z |
duke@435 | 929 | __ delayed()->restore(); |
duke@435 | 930 | #endif |
duke@435 | 931 | |
duke@435 | 932 | __ BIND(miss); |
duke@435 | 933 | __ addcc(G0,1,Rret); // set NZ flags, NZ result |
duke@435 | 934 | |
duke@435 | 935 | #if defined(COMPILER2) && !defined(_LP64) |
duke@435 | 936 | __ ld_ptr(SP,(frame::register_save_words+0)*wordSize,L0); |
duke@435 | 937 | __ ld_ptr(SP,(frame::register_save_words+1)*wordSize,L1); |
duke@435 | 938 | __ ld_ptr(SP,(frame::register_save_words+2)*wordSize,L2); |
duke@435 | 939 | __ ld_ptr(SP,(frame::register_save_words+3)*wordSize,L3); |
duke@435 | 940 | __ retl(); // Result in Rret is != 0; flags set to NZ |
duke@435 | 941 | __ delayed()->add(SP,4*wordSize,SP); |
duke@435 | 942 | #else |
duke@435 | 943 | __ ret(); // Result in Rret is != 0; flags set to NZ |
duke@435 | 944 | __ delayed()->restore(); |
duke@435 | 945 | #endif |
duke@435 | 946 | |
duke@435 | 947 | return start; |
duke@435 | 948 | } |
duke@435 | 949 | |
duke@435 | 950 | |
duke@435 | 951 | // Called from MacroAssembler::verify_oop |
duke@435 | 952 | // |
duke@435 | 953 | address generate_verify_oop_subroutine() { |
duke@435 | 954 | StubCodeMark mark(this, "StubRoutines", "verify_oop_stub"); |
duke@435 | 955 | |
duke@435 | 956 | address start = __ pc(); |
duke@435 | 957 | |
duke@435 | 958 | __ verify_oop_subroutine(); |
duke@435 | 959 | |
duke@435 | 960 | return start; |
duke@435 | 961 | } |
duke@435 | 962 | |
duke@435 | 963 | |
duke@435 | 964 | // |
duke@435 | 965 | // Verify that a register contains clean 32-bits positive value |
duke@435 | 966 | // (high 32-bits are 0) so it could be used in 64-bits shifts (sllx, srax). |
duke@435 | 967 | // |
duke@435 | 968 | // Input: |
duke@435 | 969 | // Rint - 32-bits value |
duke@435 | 970 | // Rtmp - scratch |
duke@435 | 971 | // |
duke@435 | 972 | void assert_clean_int(Register Rint, Register Rtmp) { |
duke@435 | 973 | #if defined(ASSERT) && defined(_LP64) |
duke@435 | 974 | __ signx(Rint, Rtmp); |
duke@435 | 975 | __ cmp(Rint, Rtmp); |
duke@435 | 976 | __ breakpoint_trap(Assembler::notEqual, Assembler::xcc); |
duke@435 | 977 | #endif |
duke@435 | 978 | } |
duke@435 | 979 | |
duke@435 | 980 | // |
duke@435 | 981 | // Generate overlap test for array copy stubs |
duke@435 | 982 | // |
duke@435 | 983 | // Input: |
duke@435 | 984 | // O0 - array1 |
duke@435 | 985 | // O1 - array2 |
duke@435 | 986 | // O2 - element count |
duke@435 | 987 | // |
duke@435 | 988 | // Kills temps: O3, O4 |
duke@435 | 989 | // |
duke@435 | 990 | void array_overlap_test(address no_overlap_target, int log2_elem_size) { |
duke@435 | 991 | assert(no_overlap_target != NULL, "must be generated"); |
duke@435 | 992 | array_overlap_test(no_overlap_target, NULL, log2_elem_size); |
duke@435 | 993 | } |
duke@435 | 994 | void array_overlap_test(Label& L_no_overlap, int log2_elem_size) { |
duke@435 | 995 | array_overlap_test(NULL, &L_no_overlap, log2_elem_size); |
duke@435 | 996 | } |
duke@435 | 997 | void array_overlap_test(address no_overlap_target, Label* NOLp, int log2_elem_size) { |
duke@435 | 998 | const Register from = O0; |
duke@435 | 999 | const Register to = O1; |
duke@435 | 1000 | const Register count = O2; |
duke@435 | 1001 | const Register to_from = O3; // to - from |
duke@435 | 1002 | const Register byte_count = O4; // count << log2_elem_size |
duke@435 | 1003 | |
duke@435 | 1004 | __ subcc(to, from, to_from); |
duke@435 | 1005 | __ sll_ptr(count, log2_elem_size, byte_count); |
duke@435 | 1006 | if (NOLp == NULL) |
duke@435 | 1007 | __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, no_overlap_target); |
duke@435 | 1008 | else |
duke@435 | 1009 | __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, (*NOLp)); |
duke@435 | 1010 | __ delayed()->cmp(to_from, byte_count); |
duke@435 | 1011 | if (NOLp == NULL) |
tonyp@2010 | 1012 | __ brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, no_overlap_target); |
duke@435 | 1013 | else |
tonyp@2010 | 1014 | __ brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, (*NOLp)); |
duke@435 | 1015 | __ delayed()->nop(); |
duke@435 | 1016 | } |
duke@435 | 1017 | |
duke@435 | 1018 | // |
duke@435 | 1019 | // Generate pre-write barrier for array. |
duke@435 | 1020 | // |
duke@435 | 1021 | // Input: |
duke@435 | 1022 | // addr - register containing starting address |
duke@435 | 1023 | // count - register containing element count |
duke@435 | 1024 | // tmp - scratch register |
duke@435 | 1025 | // |
duke@435 | 1026 | // The input registers are overwritten. |
duke@435 | 1027 | // |
iveresov@2606 | 1028 | void gen_write_ref_array_pre_barrier(Register addr, Register count, bool dest_uninitialized) { |
duke@435 | 1029 | BarrierSet* bs = Universe::heap()->barrier_set(); |
iveresov@2606 | 1030 | switch (bs->kind()) { |
iveresov@2606 | 1031 | case BarrierSet::G1SATBCT: |
iveresov@2606 | 1032 | case BarrierSet::G1SATBCTLogging: |
iveresov@2606 | 1033 | // With G1, don't generate the call if we statically know that the target in uninitialized |
iveresov@2606 | 1034 | if (!dest_uninitialized) { |
iveresov@2606 | 1035 | __ save_frame(0); |
iveresov@2606 | 1036 | // Save the necessary global regs... will be used after. |
iveresov@2606 | 1037 | if (addr->is_global()) { |
iveresov@2606 | 1038 | __ mov(addr, L0); |
iveresov@2606 | 1039 | } |
iveresov@2606 | 1040 | if (count->is_global()) { |
iveresov@2606 | 1041 | __ mov(count, L1); |
iveresov@2606 | 1042 | } |
iveresov@2606 | 1043 | __ mov(addr->after_save(), O0); |
iveresov@2606 | 1044 | // Get the count into O1 |
iveresov@2606 | 1045 | __ call(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre)); |
iveresov@2606 | 1046 | __ delayed()->mov(count->after_save(), O1); |
iveresov@2606 | 1047 | if (addr->is_global()) { |
iveresov@2606 | 1048 | __ mov(L0, addr); |
iveresov@2606 | 1049 | } |
iveresov@2606 | 1050 | if (count->is_global()) { |
iveresov@2606 | 1051 | __ mov(L1, count); |
iveresov@2606 | 1052 | } |
iveresov@2606 | 1053 | __ restore(); |
iveresov@2606 | 1054 | } |
iveresov@2606 | 1055 | break; |
iveresov@2606 | 1056 | case BarrierSet::CardTableModRef: |
iveresov@2606 | 1057 | case BarrierSet::CardTableExtension: |
iveresov@2606 | 1058 | case BarrierSet::ModRef: |
iveresov@2606 | 1059 | break; |
iveresov@2606 | 1060 | default: |
iveresov@2606 | 1061 | ShouldNotReachHere(); |
duke@435 | 1062 | } |
duke@435 | 1063 | } |
duke@435 | 1064 | // |
duke@435 | 1065 | // Generate post-write barrier for array. |
duke@435 | 1066 | // |
duke@435 | 1067 | // Input: |
duke@435 | 1068 | // addr - register containing starting address |
duke@435 | 1069 | // count - register containing element count |
duke@435 | 1070 | // tmp - scratch register |
duke@435 | 1071 | // |
duke@435 | 1072 | // The input registers are overwritten. |
duke@435 | 1073 | // |
duke@435 | 1074 | void gen_write_ref_array_post_barrier(Register addr, Register count, |
iveresov@2606 | 1075 | Register tmp) { |
duke@435 | 1076 | BarrierSet* bs = Universe::heap()->barrier_set(); |
duke@435 | 1077 | |
duke@435 | 1078 | switch (bs->kind()) { |
duke@435 | 1079 | case BarrierSet::G1SATBCT: |
duke@435 | 1080 | case BarrierSet::G1SATBCTLogging: |
duke@435 | 1081 | { |
duke@435 | 1082 | // Get some new fresh output registers. |
duke@435 | 1083 | __ save_frame(0); |
ysr@777 | 1084 | __ mov(addr->after_save(), O0); |
duke@435 | 1085 | __ call(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post)); |
ysr@777 | 1086 | __ delayed()->mov(count->after_save(), O1); |
duke@435 | 1087 | __ restore(); |
duke@435 | 1088 | } |
duke@435 | 1089 | break; |
duke@435 | 1090 | case BarrierSet::CardTableModRef: |
duke@435 | 1091 | case BarrierSet::CardTableExtension: |
duke@435 | 1092 | { |
duke@435 | 1093 | CardTableModRefBS* ct = (CardTableModRefBS*)bs; |
duke@435 | 1094 | assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); |
duke@435 | 1095 | assert_different_registers(addr, count, tmp); |
duke@435 | 1096 | |
duke@435 | 1097 | Label L_loop; |
duke@435 | 1098 | |
coleenp@548 | 1099 | __ sll_ptr(count, LogBytesPerHeapOop, count); |
coleenp@548 | 1100 | __ sub(count, BytesPerHeapOop, count); |
duke@435 | 1101 | __ add(count, addr, count); |
duke@435 | 1102 | // Use two shifts to clear out those low order two bits! (Cannot opt. into 1.) |
duke@435 | 1103 | __ srl_ptr(addr, CardTableModRefBS::card_shift, addr); |
duke@435 | 1104 | __ srl_ptr(count, CardTableModRefBS::card_shift, count); |
duke@435 | 1105 | __ sub(count, addr, count); |
twisti@1162 | 1106 | AddressLiteral rs(ct->byte_map_base); |
twisti@1162 | 1107 | __ set(rs, tmp); |
duke@435 | 1108 | __ BIND(L_loop); |
twisti@1162 | 1109 | __ stb(G0, tmp, addr); |
duke@435 | 1110 | __ subcc(count, 1, count); |
duke@435 | 1111 | __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop); |
duke@435 | 1112 | __ delayed()->add(addr, 1, addr); |
twisti@1162 | 1113 | } |
duke@435 | 1114 | break; |
duke@435 | 1115 | case BarrierSet::ModRef: |
duke@435 | 1116 | break; |
twisti@1162 | 1117 | default: |
duke@435 | 1118 | ShouldNotReachHere(); |
duke@435 | 1119 | } |
duke@435 | 1120 | } |
duke@435 | 1121 | |
kvn@3103 | 1122 | // |
kvn@3103 | 1123 | // Generate main code for disjoint arraycopy |
kvn@3103 | 1124 | // |
kvn@3103 | 1125 | typedef void (StubGenerator::*CopyLoopFunc)(Register from, Register to, Register count, int count_dec, |
kvn@3103 | 1126 | Label& L_loop, bool use_prefetch, bool use_bis); |
kvn@3103 | 1127 | |
kvn@3103 | 1128 | void disjoint_copy_core(Register from, Register to, Register count, int log2_elem_size, |
kvn@3103 | 1129 | int iter_size, CopyLoopFunc copy_loop_func) { |
kvn@3103 | 1130 | Label L_copy; |
kvn@3103 | 1131 | |
kvn@3103 | 1132 | assert(log2_elem_size <= 3, "the following code should be changed"); |
kvn@3103 | 1133 | int count_dec = 16>>log2_elem_size; |
kvn@3103 | 1134 | |
kvn@3103 | 1135 | int prefetch_dist = MAX2(ArraycopySrcPrefetchDistance, ArraycopyDstPrefetchDistance); |
kvn@3103 | 1136 | assert(prefetch_dist < 4096, "invalid value"); |
kvn@3103 | 1137 | prefetch_dist = (prefetch_dist + (iter_size-1)) & (-iter_size); // round up to one iteration copy size |
kvn@3103 | 1138 | int prefetch_count = (prefetch_dist >> log2_elem_size); // elements count |
kvn@3103 | 1139 | |
kvn@3103 | 1140 | if (UseBlockCopy) { |
kvn@3103 | 1141 | Label L_block_copy, L_block_copy_prefetch, L_skip_block_copy; |
kvn@3103 | 1142 | |
kvn@3103 | 1143 | // 64 bytes tail + bytes copied in one loop iteration |
kvn@3103 | 1144 | int tail_size = 64 + iter_size; |
kvn@3103 | 1145 | int block_copy_count = (MAX2(tail_size, (int)BlockCopyLowLimit)) >> log2_elem_size; |
kvn@3103 | 1146 | // Use BIS copy only for big arrays since it requires membar. |
kvn@3103 | 1147 | __ set(block_copy_count, O4); |
kvn@3103 | 1148 | __ cmp_and_br_short(count, O4, Assembler::lessUnsigned, Assembler::pt, L_skip_block_copy); |
kvn@3103 | 1149 | // This code is for disjoint source and destination: |
kvn@3103 | 1150 | // to <= from || to >= from+count |
kvn@3103 | 1151 | // but BIS will stomp over 'from' if (to > from-tail_size && to <= from) |
kvn@3103 | 1152 | __ sub(from, to, O4); |
kvn@3103 | 1153 | __ srax(O4, 4, O4); // divide by 16 since following short branch have only 5 bits for imm. |
kvn@3103 | 1154 | __ cmp_and_br_short(O4, (tail_size>>4), Assembler::lessEqualUnsigned, Assembler::pn, L_skip_block_copy); |
kvn@3103 | 1155 | |
kvn@3103 | 1156 | __ wrasi(G0, Assembler::ASI_ST_BLKINIT_PRIMARY); |
kvn@3103 | 1157 | // BIS should not be used to copy tail (64 bytes+iter_size) |
kvn@3103 | 1158 | // to avoid zeroing of following values. |
kvn@3103 | 1159 | __ sub(count, (tail_size>>log2_elem_size), count); // count is still positive >= 0 |
kvn@3103 | 1160 | |
kvn@3103 | 1161 | if (prefetch_count > 0) { // rounded up to one iteration count |
kvn@3103 | 1162 | // Do prefetching only if copy size is bigger |
kvn@3103 | 1163 | // than prefetch distance. |
kvn@3103 | 1164 | __ set(prefetch_count, O4); |
kvn@3103 | 1165 | __ cmp_and_brx_short(count, O4, Assembler::less, Assembler::pt, L_block_copy); |
kvn@3103 | 1166 | __ sub(count, prefetch_count, count); |
kvn@3103 | 1167 | |
kvn@3103 | 1168 | (this->*copy_loop_func)(from, to, count, count_dec, L_block_copy_prefetch, true, true); |
kvn@3103 | 1169 | __ add(count, prefetch_count, count); // restore count |
kvn@3103 | 1170 | |
kvn@3103 | 1171 | } // prefetch_count > 0 |
kvn@3103 | 1172 | |
kvn@3103 | 1173 | (this->*copy_loop_func)(from, to, count, count_dec, L_block_copy, false, true); |
kvn@3103 | 1174 | __ add(count, (tail_size>>log2_elem_size), count); // restore count |
kvn@3103 | 1175 | |
kvn@3103 | 1176 | __ wrasi(G0, Assembler::ASI_PRIMARY_NOFAULT); |
kvn@3103 | 1177 | // BIS needs membar. |
kvn@3103 | 1178 | __ membar(Assembler::StoreLoad); |
kvn@3103 | 1179 | // Copy tail |
kvn@3103 | 1180 | __ ba_short(L_copy); |
kvn@3103 | 1181 | |
kvn@3103 | 1182 | __ BIND(L_skip_block_copy); |
kvn@3103 | 1183 | } // UseBlockCopy |
kvn@3103 | 1184 | |
kvn@3103 | 1185 | if (prefetch_count > 0) { // rounded up to one iteration count |
kvn@3103 | 1186 | // Do prefetching only if copy size is bigger |
kvn@3103 | 1187 | // than prefetch distance. |
kvn@3103 | 1188 | __ set(prefetch_count, O4); |
kvn@3103 | 1189 | __ cmp_and_brx_short(count, O4, Assembler::lessUnsigned, Assembler::pt, L_copy); |
kvn@3103 | 1190 | __ sub(count, prefetch_count, count); |
kvn@3103 | 1191 | |
kvn@3103 | 1192 | Label L_copy_prefetch; |
kvn@3103 | 1193 | (this->*copy_loop_func)(from, to, count, count_dec, L_copy_prefetch, true, false); |
kvn@3103 | 1194 | __ add(count, prefetch_count, count); // restore count |
kvn@3103 | 1195 | |
kvn@3103 | 1196 | } // prefetch_count > 0 |
kvn@3103 | 1197 | |
kvn@3103 | 1198 | (this->*copy_loop_func)(from, to, count, count_dec, L_copy, false, false); |
kvn@3103 | 1199 | } |
kvn@3103 | 1200 | |
kvn@3103 | 1201 | |
kvn@3103 | 1202 | |
kvn@3103 | 1203 | // |
kvn@3103 | 1204 | // Helper methods for copy_16_bytes_forward_with_shift() |
kvn@3103 | 1205 | // |
kvn@3103 | 1206 | void copy_16_bytes_shift_loop(Register from, Register to, Register count, int count_dec, |
kvn@3103 | 1207 | Label& L_loop, bool use_prefetch, bool use_bis) { |
kvn@3103 | 1208 | |
kvn@3103 | 1209 | const Register left_shift = G1; // left shift bit counter |
kvn@3103 | 1210 | const Register right_shift = G5; // right shift bit counter |
kvn@3103 | 1211 | |
kvn@3103 | 1212 | __ align(OptoLoopAlignment); |
kvn@3103 | 1213 | __ BIND(L_loop); |
kvn@3103 | 1214 | if (use_prefetch) { |
kvn@3103 | 1215 | if (ArraycopySrcPrefetchDistance > 0) { |
kvn@3103 | 1216 | __ prefetch(from, ArraycopySrcPrefetchDistance, Assembler::severalReads); |
kvn@3103 | 1217 | } |
kvn@3103 | 1218 | if (ArraycopyDstPrefetchDistance > 0) { |
kvn@3103 | 1219 | __ prefetch(to, ArraycopyDstPrefetchDistance, Assembler::severalWritesAndPossiblyReads); |
kvn@3103 | 1220 | } |
kvn@3103 | 1221 | } |
kvn@3103 | 1222 | __ ldx(from, 0, O4); |
kvn@3103 | 1223 | __ ldx(from, 8, G4); |
kvn@3103 | 1224 | __ inc(to, 16); |
kvn@3103 | 1225 | __ inc(from, 16); |
kvn@3103 | 1226 | __ deccc(count, count_dec); // Can we do next iteration after this one? |
kvn@3103 | 1227 | __ srlx(O4, right_shift, G3); |
kvn@3103 | 1228 | __ bset(G3, O3); |
kvn@3103 | 1229 | __ sllx(O4, left_shift, O4); |
kvn@3103 | 1230 | __ srlx(G4, right_shift, G3); |
kvn@3103 | 1231 | __ bset(G3, O4); |
kvn@3103 | 1232 | if (use_bis) { |
kvn@3103 | 1233 | __ stxa(O3, to, -16); |
kvn@3103 | 1234 | __ stxa(O4, to, -8); |
kvn@3103 | 1235 | } else { |
kvn@3103 | 1236 | __ stx(O3, to, -16); |
kvn@3103 | 1237 | __ stx(O4, to, -8); |
kvn@3103 | 1238 | } |
kvn@3103 | 1239 | __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop); |
kvn@3103 | 1240 | __ delayed()->sllx(G4, left_shift, O3); |
kvn@3103 | 1241 | } |
duke@435 | 1242 | |
duke@435 | 1243 | // Copy big chunks forward with shift |
duke@435 | 1244 | // |
duke@435 | 1245 | // Inputs: |
duke@435 | 1246 | // from - source arrays |
duke@435 | 1247 | // to - destination array aligned to 8-bytes |
duke@435 | 1248 | // count - elements count to copy >= the count equivalent to 16 bytes |
duke@435 | 1249 | // count_dec - elements count's decrement equivalent to 16 bytes |
duke@435 | 1250 | // L_copy_bytes - copy exit label |
duke@435 | 1251 | // |
duke@435 | 1252 | void copy_16_bytes_forward_with_shift(Register from, Register to, |
kvn@3103 | 1253 | Register count, int log2_elem_size, Label& L_copy_bytes) { |
kvn@3103 | 1254 | Label L_aligned_copy, L_copy_last_bytes; |
kvn@3103 | 1255 | assert(log2_elem_size <= 3, "the following code should be changed"); |
kvn@3103 | 1256 | int count_dec = 16>>log2_elem_size; |
duke@435 | 1257 | |
duke@435 | 1258 | // if both arrays have the same alignment mod 8, do 8 bytes aligned copy |
kvn@3103 | 1259 | __ andcc(from, 7, G1); // misaligned bytes |
kvn@3103 | 1260 | __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy); |
kvn@3103 | 1261 | __ delayed()->nop(); |
duke@435 | 1262 | |
duke@435 | 1263 | const Register left_shift = G1; // left shift bit counter |
duke@435 | 1264 | const Register right_shift = G5; // right shift bit counter |
duke@435 | 1265 | |
kvn@3103 | 1266 | __ sll(G1, LogBitsPerByte, left_shift); |
kvn@3103 | 1267 | __ mov(64, right_shift); |
kvn@3103 | 1268 | __ sub(right_shift, left_shift, right_shift); |
duke@435 | 1269 | |
duke@435 | 1270 | // |
duke@435 | 1271 | // Load 2 aligned 8-bytes chunks and use one from previous iteration |
duke@435 | 1272 | // to form 2 aligned 8-bytes chunks to store. |
duke@435 | 1273 | // |
kvn@3103 | 1274 | __ dec(count, count_dec); // Pre-decrement 'count' |
kvn@3103 | 1275 | __ andn(from, 7, from); // Align address |
kvn@3103 | 1276 | __ ldx(from, 0, O3); |
kvn@3103 | 1277 | __ inc(from, 8); |
kvn@3103 | 1278 | __ sllx(O3, left_shift, O3); |
kvn@3103 | 1279 | |
kvn@3103 | 1280 | disjoint_copy_core(from, to, count, log2_elem_size, 16, copy_16_bytes_shift_loop); |
kvn@3103 | 1281 | |
kvn@3103 | 1282 | __ inccc(count, count_dec>>1 ); // + 8 bytes |
kvn@3103 | 1283 | __ brx(Assembler::negative, true, Assembler::pn, L_copy_last_bytes); |
kvn@3103 | 1284 | __ delayed()->inc(count, count_dec>>1); // restore 'count' |
kvn@3103 | 1285 | |
kvn@3103 | 1286 | // copy 8 bytes, part of them already loaded in O3 |
kvn@3103 | 1287 | __ ldx(from, 0, O4); |
kvn@3103 | 1288 | __ inc(to, 8); |
kvn@3103 | 1289 | __ inc(from, 8); |
kvn@3103 | 1290 | __ srlx(O4, right_shift, G3); |
kvn@3103 | 1291 | __ bset(O3, G3); |
kvn@3103 | 1292 | __ stx(G3, to, -8); |
duke@435 | 1293 | |
duke@435 | 1294 | __ BIND(L_copy_last_bytes); |
kvn@3103 | 1295 | __ srl(right_shift, LogBitsPerByte, right_shift); // misaligned bytes |
kvn@3103 | 1296 | __ br(Assembler::always, false, Assembler::pt, L_copy_bytes); |
kvn@3103 | 1297 | __ delayed()->sub(from, right_shift, from); // restore address |
duke@435 | 1298 | |
duke@435 | 1299 | __ BIND(L_aligned_copy); |
duke@435 | 1300 | } |
duke@435 | 1301 | |
duke@435 | 1302 | // Copy big chunks backward with shift |
duke@435 | 1303 | // |
duke@435 | 1304 | // Inputs: |
duke@435 | 1305 | // end_from - source arrays end address |
duke@435 | 1306 | // end_to - destination array end address aligned to 8-bytes |
duke@435 | 1307 | // count - elements count to copy >= the count equivalent to 16 bytes |
duke@435 | 1308 | // count_dec - elements count's decrement equivalent to 16 bytes |
duke@435 | 1309 | // L_aligned_copy - aligned copy exit label |
duke@435 | 1310 | // L_copy_bytes - copy exit label |
duke@435 | 1311 | // |
duke@435 | 1312 | void copy_16_bytes_backward_with_shift(Register end_from, Register end_to, |
duke@435 | 1313 | Register count, int count_dec, |
duke@435 | 1314 | Label& L_aligned_copy, Label& L_copy_bytes) { |
duke@435 | 1315 | Label L_loop, L_copy_last_bytes; |
duke@435 | 1316 | |
duke@435 | 1317 | // if both arrays have the same alignment mod 8, do 8 bytes aligned copy |
duke@435 | 1318 | __ andcc(end_from, 7, G1); // misaligned bytes |
duke@435 | 1319 | __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy); |
duke@435 | 1320 | __ delayed()->deccc(count, count_dec); // Pre-decrement 'count' |
duke@435 | 1321 | |
duke@435 | 1322 | const Register left_shift = G1; // left shift bit counter |
duke@435 | 1323 | const Register right_shift = G5; // right shift bit counter |
duke@435 | 1324 | |
duke@435 | 1325 | __ sll(G1, LogBitsPerByte, left_shift); |
duke@435 | 1326 | __ mov(64, right_shift); |
duke@435 | 1327 | __ sub(right_shift, left_shift, right_shift); |
duke@435 | 1328 | |
duke@435 | 1329 | // |
duke@435 | 1330 | // Load 2 aligned 8-bytes chunks and use one from previous iteration |
duke@435 | 1331 | // to form 2 aligned 8-bytes chunks to store. |
duke@435 | 1332 | // |
duke@435 | 1333 | __ andn(end_from, 7, end_from); // Align address |
duke@435 | 1334 | __ ldx(end_from, 0, O3); |
kvn@1800 | 1335 | __ align(OptoLoopAlignment); |
duke@435 | 1336 | __ BIND(L_loop); |
duke@435 | 1337 | __ ldx(end_from, -8, O4); |
duke@435 | 1338 | __ deccc(count, count_dec); // Can we do next iteration after this one? |
duke@435 | 1339 | __ ldx(end_from, -16, G4); |
duke@435 | 1340 | __ dec(end_to, 16); |
duke@435 | 1341 | __ dec(end_from, 16); |
duke@435 | 1342 | __ srlx(O3, right_shift, O3); |
duke@435 | 1343 | __ sllx(O4, left_shift, G3); |
duke@435 | 1344 | __ bset(G3, O3); |
duke@435 | 1345 | __ stx(O3, end_to, 8); |
duke@435 | 1346 | __ srlx(O4, right_shift, O4); |
duke@435 | 1347 | __ sllx(G4, left_shift, G3); |
duke@435 | 1348 | __ bset(G3, O4); |
duke@435 | 1349 | __ stx(O4, end_to, 0); |
duke@435 | 1350 | __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop); |
duke@435 | 1351 | __ delayed()->mov(G4, O3); |
duke@435 | 1352 | |
duke@435 | 1353 | __ inccc(count, count_dec>>1 ); // + 8 bytes |
duke@435 | 1354 | __ brx(Assembler::negative, true, Assembler::pn, L_copy_last_bytes); |
duke@435 | 1355 | __ delayed()->inc(count, count_dec>>1); // restore 'count' |
duke@435 | 1356 | |
duke@435 | 1357 | // copy 8 bytes, part of them already loaded in O3 |
duke@435 | 1358 | __ ldx(end_from, -8, O4); |
duke@435 | 1359 | __ dec(end_to, 8); |
duke@435 | 1360 | __ dec(end_from, 8); |
duke@435 | 1361 | __ srlx(O3, right_shift, O3); |
duke@435 | 1362 | __ sllx(O4, left_shift, G3); |
duke@435 | 1363 | __ bset(O3, G3); |
duke@435 | 1364 | __ stx(G3, end_to, 0); |
duke@435 | 1365 | |
duke@435 | 1366 | __ BIND(L_copy_last_bytes); |
duke@435 | 1367 | __ srl(left_shift, LogBitsPerByte, left_shift); // misaligned bytes |
duke@435 | 1368 | __ br(Assembler::always, false, Assembler::pt, L_copy_bytes); |
duke@435 | 1369 | __ delayed()->add(end_from, left_shift, end_from); // restore address |
duke@435 | 1370 | } |
duke@435 | 1371 | |
duke@435 | 1372 | // |
duke@435 | 1373 | // Generate stub for disjoint byte copy. If "aligned" is true, the |
duke@435 | 1374 | // "from" and "to" addresses are assumed to be heapword aligned. |
duke@435 | 1375 | // |
duke@435 | 1376 | // Arguments for generated stub: |
duke@435 | 1377 | // from: O0 |
duke@435 | 1378 | // to: O1 |
duke@435 | 1379 | // count: O2 treated as signed |
duke@435 | 1380 | // |
iveresov@2595 | 1381 | address generate_disjoint_byte_copy(bool aligned, address *entry, const char *name) { |
duke@435 | 1382 | __ align(CodeEntryAlignment); |
duke@435 | 1383 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1384 | address start = __ pc(); |
duke@435 | 1385 | |
duke@435 | 1386 | Label L_skip_alignment, L_align; |
duke@435 | 1387 | Label L_copy_byte, L_copy_byte_loop, L_exit; |
duke@435 | 1388 | |
duke@435 | 1389 | const Register from = O0; // source array address |
duke@435 | 1390 | const Register to = O1; // destination array address |
duke@435 | 1391 | const Register count = O2; // elements count |
duke@435 | 1392 | const Register offset = O5; // offset from start of arrays |
duke@435 | 1393 | // O3, O4, G3, G4 are used as temp registers |
duke@435 | 1394 | |
duke@435 | 1395 | assert_clean_int(count, O3); // Make sure 'count' is clean int. |
duke@435 | 1396 | |
iveresov@2595 | 1397 | if (entry != NULL) { |
iveresov@2595 | 1398 | *entry = __ pc(); |
iveresov@2595 | 1399 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1400 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1401 | } |
duke@435 | 1402 | |
duke@435 | 1403 | // for short arrays, just do single element copy |
duke@435 | 1404 | __ cmp(count, 23); // 16 + 7 |
duke@435 | 1405 | __ brx(Assembler::less, false, Assembler::pn, L_copy_byte); |
duke@435 | 1406 | __ delayed()->mov(G0, offset); |
duke@435 | 1407 | |
duke@435 | 1408 | if (aligned) { |
duke@435 | 1409 | // 'aligned' == true when it is known statically during compilation |
duke@435 | 1410 | // of this arraycopy call site that both 'from' and 'to' addresses |
duke@435 | 1411 | // are HeapWordSize aligned (see LibraryCallKit::basictype2arraycopy()). |
duke@435 | 1412 | // |
duke@435 | 1413 | // Aligned arrays have 4 bytes alignment in 32-bits VM |
duke@435 | 1414 | // and 8 bytes - in 64-bits VM. So we do it only for 32-bits VM |
duke@435 | 1415 | // |
duke@435 | 1416 | #ifndef _LP64 |
duke@435 | 1417 | // copy a 4-bytes word if necessary to align 'to' to 8 bytes |
duke@435 | 1418 | __ andcc(to, 7, G0); |
duke@435 | 1419 | __ br(Assembler::zero, false, Assembler::pn, L_skip_alignment); |
duke@435 | 1420 | __ delayed()->ld(from, 0, O3); |
duke@435 | 1421 | __ inc(from, 4); |
duke@435 | 1422 | __ inc(to, 4); |
duke@435 | 1423 | __ dec(count, 4); |
duke@435 | 1424 | __ st(O3, to, -4); |
duke@435 | 1425 | __ BIND(L_skip_alignment); |
duke@435 | 1426 | #endif |
duke@435 | 1427 | } else { |
duke@435 | 1428 | // copy bytes to align 'to' on 8 byte boundary |
duke@435 | 1429 | __ andcc(to, 7, G1); // misaligned bytes |
duke@435 | 1430 | __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); |
duke@435 | 1431 | __ delayed()->neg(G1); |
duke@435 | 1432 | __ inc(G1, 8); // bytes need to copy to next 8-bytes alignment |
duke@435 | 1433 | __ sub(count, G1, count); |
duke@435 | 1434 | __ BIND(L_align); |
duke@435 | 1435 | __ ldub(from, 0, O3); |
duke@435 | 1436 | __ deccc(G1); |
duke@435 | 1437 | __ inc(from); |
duke@435 | 1438 | __ stb(O3, to, 0); |
duke@435 | 1439 | __ br(Assembler::notZero, false, Assembler::pt, L_align); |
duke@435 | 1440 | __ delayed()->inc(to); |
duke@435 | 1441 | __ BIND(L_skip_alignment); |
duke@435 | 1442 | } |
duke@435 | 1443 | #ifdef _LP64 |
duke@435 | 1444 | if (!aligned) |
duke@435 | 1445 | #endif |
duke@435 | 1446 | { |
duke@435 | 1447 | // Copy with shift 16 bytes per iteration if arrays do not have |
duke@435 | 1448 | // the same alignment mod 8, otherwise fall through to the next |
duke@435 | 1449 | // code for aligned copy. |
duke@435 | 1450 | // The compare above (count >= 23) guarantes 'count' >= 16 bytes. |
duke@435 | 1451 | // Also jump over aligned copy after the copy with shift completed. |
duke@435 | 1452 | |
kvn@3103 | 1453 | copy_16_bytes_forward_with_shift(from, to, count, 0, L_copy_byte); |
duke@435 | 1454 | } |
duke@435 | 1455 | |
duke@435 | 1456 | // Both array are 8 bytes aligned, copy 16 bytes at a time |
duke@435 | 1457 | __ and3(count, 7, G4); // Save count |
duke@435 | 1458 | __ srl(count, 3, count); |
duke@435 | 1459 | generate_disjoint_long_copy_core(aligned); |
duke@435 | 1460 | __ mov(G4, count); // Restore count |
duke@435 | 1461 | |
duke@435 | 1462 | // copy tailing bytes |
duke@435 | 1463 | __ BIND(L_copy_byte); |
kvn@3037 | 1464 | __ cmp_and_br_short(count, 0, Assembler::equal, Assembler::pt, L_exit); |
kvn@1800 | 1465 | __ align(OptoLoopAlignment); |
duke@435 | 1466 | __ BIND(L_copy_byte_loop); |
duke@435 | 1467 | __ ldub(from, offset, O3); |
duke@435 | 1468 | __ deccc(count); |
duke@435 | 1469 | __ stb(O3, to, offset); |
duke@435 | 1470 | __ brx(Assembler::notZero, false, Assembler::pt, L_copy_byte_loop); |
duke@435 | 1471 | __ delayed()->inc(offset); |
duke@435 | 1472 | |
duke@435 | 1473 | __ BIND(L_exit); |
duke@435 | 1474 | // O3, O4 are used as temp registers |
duke@435 | 1475 | inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr, O3, O4); |
duke@435 | 1476 | __ retl(); |
duke@435 | 1477 | __ delayed()->mov(G0, O0); // return 0 |
duke@435 | 1478 | return start; |
duke@435 | 1479 | } |
duke@435 | 1480 | |
duke@435 | 1481 | // |
duke@435 | 1482 | // Generate stub for conjoint byte copy. If "aligned" is true, the |
duke@435 | 1483 | // "from" and "to" addresses are assumed to be heapword aligned. |
duke@435 | 1484 | // |
duke@435 | 1485 | // Arguments for generated stub: |
duke@435 | 1486 | // from: O0 |
duke@435 | 1487 | // to: O1 |
duke@435 | 1488 | // count: O2 treated as signed |
duke@435 | 1489 | // |
iveresov@2595 | 1490 | address generate_conjoint_byte_copy(bool aligned, address nooverlap_target, |
iveresov@2595 | 1491 | address *entry, const char *name) { |
duke@435 | 1492 | // Do reverse copy. |
duke@435 | 1493 | |
duke@435 | 1494 | __ align(CodeEntryAlignment); |
duke@435 | 1495 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1496 | address start = __ pc(); |
duke@435 | 1497 | |
duke@435 | 1498 | Label L_skip_alignment, L_align, L_aligned_copy; |
duke@435 | 1499 | Label L_copy_byte, L_copy_byte_loop, L_exit; |
duke@435 | 1500 | |
duke@435 | 1501 | const Register from = O0; // source array address |
duke@435 | 1502 | const Register to = O1; // destination array address |
duke@435 | 1503 | const Register count = O2; // elements count |
duke@435 | 1504 | const Register end_from = from; // source array end address |
duke@435 | 1505 | const Register end_to = to; // destination array end address |
duke@435 | 1506 | |
duke@435 | 1507 | assert_clean_int(count, O3); // Make sure 'count' is clean int. |
duke@435 | 1508 | |
iveresov@2595 | 1509 | if (entry != NULL) { |
iveresov@2595 | 1510 | *entry = __ pc(); |
iveresov@2595 | 1511 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1512 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1513 | } |
duke@435 | 1514 | |
duke@435 | 1515 | array_overlap_test(nooverlap_target, 0); |
duke@435 | 1516 | |
duke@435 | 1517 | __ add(to, count, end_to); // offset after last copied element |
duke@435 | 1518 | |
duke@435 | 1519 | // for short arrays, just do single element copy |
duke@435 | 1520 | __ cmp(count, 23); // 16 + 7 |
duke@435 | 1521 | __ brx(Assembler::less, false, Assembler::pn, L_copy_byte); |
duke@435 | 1522 | __ delayed()->add(from, count, end_from); |
duke@435 | 1523 | |
duke@435 | 1524 | { |
duke@435 | 1525 | // Align end of arrays since they could be not aligned even |
duke@435 | 1526 | // when arrays itself are aligned. |
duke@435 | 1527 | |
duke@435 | 1528 | // copy bytes to align 'end_to' on 8 byte boundary |
duke@435 | 1529 | __ andcc(end_to, 7, G1); // misaligned bytes |
duke@435 | 1530 | __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); |
duke@435 | 1531 | __ delayed()->nop(); |
duke@435 | 1532 | __ sub(count, G1, count); |
duke@435 | 1533 | __ BIND(L_align); |
duke@435 | 1534 | __ dec(end_from); |
duke@435 | 1535 | __ dec(end_to); |
duke@435 | 1536 | __ ldub(end_from, 0, O3); |
duke@435 | 1537 | __ deccc(G1); |
duke@435 | 1538 | __ brx(Assembler::notZero, false, Assembler::pt, L_align); |
duke@435 | 1539 | __ delayed()->stb(O3, end_to, 0); |
duke@435 | 1540 | __ BIND(L_skip_alignment); |
duke@435 | 1541 | } |
duke@435 | 1542 | #ifdef _LP64 |
duke@435 | 1543 | if (aligned) { |
duke@435 | 1544 | // Both arrays are aligned to 8-bytes in 64-bits VM. |
duke@435 | 1545 | // The 'count' is decremented in copy_16_bytes_backward_with_shift() |
duke@435 | 1546 | // in unaligned case. |
duke@435 | 1547 | __ dec(count, 16); |
duke@435 | 1548 | } else |
duke@435 | 1549 | #endif |
duke@435 | 1550 | { |
duke@435 | 1551 | // Copy with shift 16 bytes per iteration if arrays do not have |
duke@435 | 1552 | // the same alignment mod 8, otherwise jump to the next |
duke@435 | 1553 | // code for aligned copy (and substracting 16 from 'count' before jump). |
duke@435 | 1554 | // The compare above (count >= 11) guarantes 'count' >= 16 bytes. |
duke@435 | 1555 | // Also jump over aligned copy after the copy with shift completed. |
duke@435 | 1556 | |
duke@435 | 1557 | copy_16_bytes_backward_with_shift(end_from, end_to, count, 16, |
duke@435 | 1558 | L_aligned_copy, L_copy_byte); |
duke@435 | 1559 | } |
duke@435 | 1560 | // copy 4 elements (16 bytes) at a time |
kvn@1800 | 1561 | __ align(OptoLoopAlignment); |
duke@435 | 1562 | __ BIND(L_aligned_copy); |
duke@435 | 1563 | __ dec(end_from, 16); |
duke@435 | 1564 | __ ldx(end_from, 8, O3); |
duke@435 | 1565 | __ ldx(end_from, 0, O4); |
duke@435 | 1566 | __ dec(end_to, 16); |
duke@435 | 1567 | __ deccc(count, 16); |
duke@435 | 1568 | __ stx(O3, end_to, 8); |
duke@435 | 1569 | __ brx(Assembler::greaterEqual, false, Assembler::pt, L_aligned_copy); |
duke@435 | 1570 | __ delayed()->stx(O4, end_to, 0); |
duke@435 | 1571 | __ inc(count, 16); |
duke@435 | 1572 | |
duke@435 | 1573 | // copy 1 element (2 bytes) at a time |
duke@435 | 1574 | __ BIND(L_copy_byte); |
kvn@3037 | 1575 | __ cmp_and_br_short(count, 0, Assembler::equal, Assembler::pt, L_exit); |
kvn@1800 | 1576 | __ align(OptoLoopAlignment); |
duke@435 | 1577 | __ BIND(L_copy_byte_loop); |
duke@435 | 1578 | __ dec(end_from); |
duke@435 | 1579 | __ dec(end_to); |
duke@435 | 1580 | __ ldub(end_from, 0, O4); |
duke@435 | 1581 | __ deccc(count); |
duke@435 | 1582 | __ brx(Assembler::greater, false, Assembler::pt, L_copy_byte_loop); |
duke@435 | 1583 | __ delayed()->stb(O4, end_to, 0); |
duke@435 | 1584 | |
duke@435 | 1585 | __ BIND(L_exit); |
duke@435 | 1586 | // O3, O4 are used as temp registers |
duke@435 | 1587 | inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr, O3, O4); |
duke@435 | 1588 | __ retl(); |
duke@435 | 1589 | __ delayed()->mov(G0, O0); // return 0 |
duke@435 | 1590 | return start; |
duke@435 | 1591 | } |
duke@435 | 1592 | |
duke@435 | 1593 | // |
duke@435 | 1594 | // Generate stub for disjoint short copy. If "aligned" is true, the |
duke@435 | 1595 | // "from" and "to" addresses are assumed to be heapword aligned. |
duke@435 | 1596 | // |
duke@435 | 1597 | // Arguments for generated stub: |
duke@435 | 1598 | // from: O0 |
duke@435 | 1599 | // to: O1 |
duke@435 | 1600 | // count: O2 treated as signed |
duke@435 | 1601 | // |
iveresov@2595 | 1602 | address generate_disjoint_short_copy(bool aligned, address *entry, const char * name) { |
duke@435 | 1603 | __ align(CodeEntryAlignment); |
duke@435 | 1604 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1605 | address start = __ pc(); |
duke@435 | 1606 | |
duke@435 | 1607 | Label L_skip_alignment, L_skip_alignment2; |
duke@435 | 1608 | Label L_copy_2_bytes, L_copy_2_bytes_loop, L_exit; |
duke@435 | 1609 | |
duke@435 | 1610 | const Register from = O0; // source array address |
duke@435 | 1611 | const Register to = O1; // destination array address |
duke@435 | 1612 | const Register count = O2; // elements count |
duke@435 | 1613 | const Register offset = O5; // offset from start of arrays |
duke@435 | 1614 | // O3, O4, G3, G4 are used as temp registers |
duke@435 | 1615 | |
duke@435 | 1616 | assert_clean_int(count, O3); // Make sure 'count' is clean int. |
duke@435 | 1617 | |
iveresov@2595 | 1618 | if (entry != NULL) { |
iveresov@2595 | 1619 | *entry = __ pc(); |
iveresov@2595 | 1620 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1621 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1622 | } |
duke@435 | 1623 | |
duke@435 | 1624 | // for short arrays, just do single element copy |
duke@435 | 1625 | __ cmp(count, 11); // 8 + 3 (22 bytes) |
duke@435 | 1626 | __ brx(Assembler::less, false, Assembler::pn, L_copy_2_bytes); |
duke@435 | 1627 | __ delayed()->mov(G0, offset); |
duke@435 | 1628 | |
duke@435 | 1629 | if (aligned) { |
duke@435 | 1630 | // 'aligned' == true when it is known statically during compilation |
duke@435 | 1631 | // of this arraycopy call site that both 'from' and 'to' addresses |
duke@435 | 1632 | // are HeapWordSize aligned (see LibraryCallKit::basictype2arraycopy()). |
duke@435 | 1633 | // |
duke@435 | 1634 | // Aligned arrays have 4 bytes alignment in 32-bits VM |
duke@435 | 1635 | // and 8 bytes - in 64-bits VM. |
duke@435 | 1636 | // |
duke@435 | 1637 | #ifndef _LP64 |
duke@435 | 1638 | // copy a 2-elements word if necessary to align 'to' to 8 bytes |
duke@435 | 1639 | __ andcc(to, 7, G0); |
duke@435 | 1640 | __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); |
duke@435 | 1641 | __ delayed()->ld(from, 0, O3); |
duke@435 | 1642 | __ inc(from, 4); |
duke@435 | 1643 | __ inc(to, 4); |
duke@435 | 1644 | __ dec(count, 2); |
duke@435 | 1645 | __ st(O3, to, -4); |
duke@435 | 1646 | __ BIND(L_skip_alignment); |
duke@435 | 1647 | #endif |
duke@435 | 1648 | } else { |
duke@435 | 1649 | // copy 1 element if necessary to align 'to' on an 4 bytes |
duke@435 | 1650 | __ andcc(to, 3, G0); |
duke@435 | 1651 | __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); |
duke@435 | 1652 | __ delayed()->lduh(from, 0, O3); |
duke@435 | 1653 | __ inc(from, 2); |
duke@435 | 1654 | __ inc(to, 2); |
duke@435 | 1655 | __ dec(count); |
duke@435 | 1656 | __ sth(O3, to, -2); |
duke@435 | 1657 | __ BIND(L_skip_alignment); |
duke@435 | 1658 | |
duke@435 | 1659 | // copy 2 elements to align 'to' on an 8 byte boundary |
duke@435 | 1660 | __ andcc(to, 7, G0); |
duke@435 | 1661 | __ br(Assembler::zero, false, Assembler::pn, L_skip_alignment2); |
duke@435 | 1662 | __ delayed()->lduh(from, 0, O3); |
duke@435 | 1663 | __ dec(count, 2); |
duke@435 | 1664 | __ lduh(from, 2, O4); |
duke@435 | 1665 | __ inc(from, 4); |
duke@435 | 1666 | __ inc(to, 4); |
duke@435 | 1667 | __ sth(O3, to, -4); |
duke@435 | 1668 | __ sth(O4, to, -2); |
duke@435 | 1669 | __ BIND(L_skip_alignment2); |
duke@435 | 1670 | } |
duke@435 | 1671 | #ifdef _LP64 |
duke@435 | 1672 | if (!aligned) |
duke@435 | 1673 | #endif |
duke@435 | 1674 | { |
duke@435 | 1675 | // Copy with shift 16 bytes per iteration if arrays do not have |
duke@435 | 1676 | // the same alignment mod 8, otherwise fall through to the next |
duke@435 | 1677 | // code for aligned copy. |
duke@435 | 1678 | // The compare above (count >= 11) guarantes 'count' >= 16 bytes. |
duke@435 | 1679 | // Also jump over aligned copy after the copy with shift completed. |
duke@435 | 1680 | |
kvn@3103 | 1681 | copy_16_bytes_forward_with_shift(from, to, count, 1, L_copy_2_bytes); |
duke@435 | 1682 | } |
duke@435 | 1683 | |
duke@435 | 1684 | // Both array are 8 bytes aligned, copy 16 bytes at a time |
duke@435 | 1685 | __ and3(count, 3, G4); // Save |
duke@435 | 1686 | __ srl(count, 2, count); |
duke@435 | 1687 | generate_disjoint_long_copy_core(aligned); |
duke@435 | 1688 | __ mov(G4, count); // restore |
duke@435 | 1689 | |
duke@435 | 1690 | // copy 1 element at a time |
duke@435 | 1691 | __ BIND(L_copy_2_bytes); |
kvn@3037 | 1692 | __ cmp_and_br_short(count, 0, Assembler::equal, Assembler::pt, L_exit); |
kvn@1800 | 1693 | __ align(OptoLoopAlignment); |
duke@435 | 1694 | __ BIND(L_copy_2_bytes_loop); |
duke@435 | 1695 | __ lduh(from, offset, O3); |
duke@435 | 1696 | __ deccc(count); |
duke@435 | 1697 | __ sth(O3, to, offset); |
duke@435 | 1698 | __ brx(Assembler::notZero, false, Assembler::pt, L_copy_2_bytes_loop); |
duke@435 | 1699 | __ delayed()->inc(offset, 2); |
duke@435 | 1700 | |
duke@435 | 1701 | __ BIND(L_exit); |
duke@435 | 1702 | // O3, O4 are used as temp registers |
duke@435 | 1703 | inc_counter_np(SharedRuntime::_jshort_array_copy_ctr, O3, O4); |
duke@435 | 1704 | __ retl(); |
duke@435 | 1705 | __ delayed()->mov(G0, O0); // return 0 |
duke@435 | 1706 | return start; |
duke@435 | 1707 | } |
duke@435 | 1708 | |
duke@435 | 1709 | // |
never@2118 | 1710 | // Generate stub for disjoint short fill. If "aligned" is true, the |
never@2118 | 1711 | // "to" address is assumed to be heapword aligned. |
never@2118 | 1712 | // |
never@2118 | 1713 | // Arguments for generated stub: |
never@2118 | 1714 | // to: O0 |
never@2118 | 1715 | // value: O1 |
never@2118 | 1716 | // count: O2 treated as signed |
never@2118 | 1717 | // |
never@2118 | 1718 | address generate_fill(BasicType t, bool aligned, const char* name) { |
never@2118 | 1719 | __ align(CodeEntryAlignment); |
never@2118 | 1720 | StubCodeMark mark(this, "StubRoutines", name); |
never@2118 | 1721 | address start = __ pc(); |
never@2118 | 1722 | |
never@2118 | 1723 | const Register to = O0; // source array address |
never@2118 | 1724 | const Register value = O1; // fill value |
never@2118 | 1725 | const Register count = O2; // elements count |
never@2118 | 1726 | // O3 is used as a temp register |
never@2118 | 1727 | |
never@2118 | 1728 | assert_clean_int(count, O3); // Make sure 'count' is clean int. |
never@2118 | 1729 | |
never@2118 | 1730 | Label L_exit, L_skip_align1, L_skip_align2, L_fill_byte; |
never@2149 | 1731 | Label L_fill_2_bytes, L_fill_elements, L_fill_32_bytes; |
never@2118 | 1732 | |
never@2118 | 1733 | int shift = -1; |
never@2118 | 1734 | switch (t) { |
never@2118 | 1735 | case T_BYTE: |
never@2118 | 1736 | shift = 2; |
never@2118 | 1737 | break; |
never@2118 | 1738 | case T_SHORT: |
never@2118 | 1739 | shift = 1; |
never@2118 | 1740 | break; |
never@2118 | 1741 | case T_INT: |
never@2118 | 1742 | shift = 0; |
never@2118 | 1743 | break; |
never@2118 | 1744 | default: ShouldNotReachHere(); |
never@2118 | 1745 | } |
never@2118 | 1746 | |
never@2118 | 1747 | BLOCK_COMMENT("Entry:"); |
never@2118 | 1748 | |
never@2118 | 1749 | if (t == T_BYTE) { |
never@2118 | 1750 | // Zero extend value |
never@2118 | 1751 | __ and3(value, 0xff, value); |
never@2118 | 1752 | __ sllx(value, 8, O3); |
never@2118 | 1753 | __ or3(value, O3, value); |
never@2118 | 1754 | } |
never@2118 | 1755 | if (t == T_SHORT) { |
never@2118 | 1756 | // Zero extend value |
never@2149 | 1757 | __ sllx(value, 48, value); |
never@2149 | 1758 | __ srlx(value, 48, value); |
never@2118 | 1759 | } |
never@2118 | 1760 | if (t == T_BYTE || t == T_SHORT) { |
never@2118 | 1761 | __ sllx(value, 16, O3); |
never@2118 | 1762 | __ or3(value, O3, value); |
never@2118 | 1763 | } |
never@2118 | 1764 | |
never@2118 | 1765 | __ cmp(count, 2<<shift); // Short arrays (< 8 bytes) fill by element |
never@2149 | 1766 | __ brx(Assembler::lessUnsigned, false, Assembler::pn, L_fill_elements); // use unsigned cmp |
never@2149 | 1767 | __ delayed()->andcc(count, 1, G0); |
never@2118 | 1768 | |
never@2118 | 1769 | if (!aligned && (t == T_BYTE || t == T_SHORT)) { |
never@2118 | 1770 | // align source address at 4 bytes address boundary |
never@2118 | 1771 | if (t == T_BYTE) { |
never@2118 | 1772 | // One byte misalignment happens only for byte arrays |
never@2118 | 1773 | __ andcc(to, 1, G0); |
never@2118 | 1774 | __ br(Assembler::zero, false, Assembler::pt, L_skip_align1); |
never@2118 | 1775 | __ delayed()->nop(); |
never@2118 | 1776 | __ stb(value, to, 0); |
never@2118 | 1777 | __ inc(to, 1); |
never@2118 | 1778 | __ dec(count, 1); |
never@2118 | 1779 | __ BIND(L_skip_align1); |
never@2118 | 1780 | } |
never@2118 | 1781 | // Two bytes misalignment happens only for byte and short (char) arrays |
never@2118 | 1782 | __ andcc(to, 2, G0); |
never@2118 | 1783 | __ br(Assembler::zero, false, Assembler::pt, L_skip_align2); |
never@2118 | 1784 | __ delayed()->nop(); |
never@2118 | 1785 | __ sth(value, to, 0); |
never@2118 | 1786 | __ inc(to, 2); |
never@2118 | 1787 | __ dec(count, 1 << (shift - 1)); |
never@2118 | 1788 | __ BIND(L_skip_align2); |
never@2118 | 1789 | } |
never@2118 | 1790 | #ifdef _LP64 |
never@2118 | 1791 | if (!aligned) { |
never@2118 | 1792 | #endif |
never@2118 | 1793 | // align to 8 bytes, we know we are 4 byte aligned to start |
never@2118 | 1794 | __ andcc(to, 7, G0); |
never@2118 | 1795 | __ br(Assembler::zero, false, Assembler::pt, L_fill_32_bytes); |
never@2118 | 1796 | __ delayed()->nop(); |
never@2118 | 1797 | __ stw(value, to, 0); |
never@2118 | 1798 | __ inc(to, 4); |
never@2118 | 1799 | __ dec(count, 1 << shift); |
never@2118 | 1800 | __ BIND(L_fill_32_bytes); |
never@2118 | 1801 | #ifdef _LP64 |
never@2118 | 1802 | } |
never@2118 | 1803 | #endif |
never@2118 | 1804 | |
never@2118 | 1805 | if (t == T_INT) { |
never@2118 | 1806 | // Zero extend value |
never@2118 | 1807 | __ srl(value, 0, value); |
never@2118 | 1808 | } |
never@2118 | 1809 | if (t == T_BYTE || t == T_SHORT || t == T_INT) { |
never@2118 | 1810 | __ sllx(value, 32, O3); |
never@2118 | 1811 | __ or3(value, O3, value); |
never@2118 | 1812 | } |
never@2118 | 1813 | |
never@2137 | 1814 | Label L_check_fill_8_bytes; |
never@2137 | 1815 | // Fill 32-byte chunks |
never@2137 | 1816 | __ subcc(count, 8 << shift, count); |
never@2137 | 1817 | __ brx(Assembler::less, false, Assembler::pt, L_check_fill_8_bytes); |
never@2137 | 1818 | __ delayed()->nop(); |
never@2137 | 1819 | |
never@2149 | 1820 | Label L_fill_32_bytes_loop, L_fill_4_bytes; |
never@2118 | 1821 | __ align(16); |
never@2118 | 1822 | __ BIND(L_fill_32_bytes_loop); |
never@2118 | 1823 | |
never@2118 | 1824 | __ stx(value, to, 0); |
never@2118 | 1825 | __ stx(value, to, 8); |
never@2118 | 1826 | __ stx(value, to, 16); |
never@2118 | 1827 | __ stx(value, to, 24); |
never@2118 | 1828 | |
never@2118 | 1829 | __ subcc(count, 8 << shift, count); |
never@2118 | 1830 | __ brx(Assembler::greaterEqual, false, Assembler::pt, L_fill_32_bytes_loop); |
never@2118 | 1831 | __ delayed()->add(to, 32, to); |
never@2118 | 1832 | |
never@2118 | 1833 | __ BIND(L_check_fill_8_bytes); |
never@2118 | 1834 | __ addcc(count, 8 << shift, count); |
never@2118 | 1835 | __ brx(Assembler::zero, false, Assembler::pn, L_exit); |
never@2118 | 1836 | __ delayed()->subcc(count, 1 << (shift + 1), count); |
never@2118 | 1837 | __ brx(Assembler::less, false, Assembler::pn, L_fill_4_bytes); |
never@2118 | 1838 | __ delayed()->andcc(count, 1<<shift, G0); |
never@2118 | 1839 | |
never@2118 | 1840 | // |
never@2118 | 1841 | // length is too short, just fill 8 bytes at a time |
never@2118 | 1842 | // |
never@2118 | 1843 | Label L_fill_8_bytes_loop; |
never@2118 | 1844 | __ BIND(L_fill_8_bytes_loop); |
never@2118 | 1845 | __ stx(value, to, 0); |
never@2118 | 1846 | __ subcc(count, 1 << (shift + 1), count); |
never@2118 | 1847 | __ brx(Assembler::greaterEqual, false, Assembler::pn, L_fill_8_bytes_loop); |
never@2118 | 1848 | __ delayed()->add(to, 8, to); |
never@2118 | 1849 | |
never@2118 | 1850 | // fill trailing 4 bytes |
never@2118 | 1851 | __ andcc(count, 1<<shift, G0); // in delay slot of branches |
never@2149 | 1852 | if (t == T_INT) { |
never@2149 | 1853 | __ BIND(L_fill_elements); |
never@2149 | 1854 | } |
never@2118 | 1855 | __ BIND(L_fill_4_bytes); |
never@2118 | 1856 | __ brx(Assembler::zero, false, Assembler::pt, L_fill_2_bytes); |
never@2118 | 1857 | if (t == T_BYTE || t == T_SHORT) { |
never@2118 | 1858 | __ delayed()->andcc(count, 1<<(shift-1), G0); |
never@2118 | 1859 | } else { |
never@2118 | 1860 | __ delayed()->nop(); |
never@2118 | 1861 | } |
never@2118 | 1862 | __ stw(value, to, 0); |
never@2118 | 1863 | if (t == T_BYTE || t == T_SHORT) { |
never@2118 | 1864 | __ inc(to, 4); |
never@2118 | 1865 | // fill trailing 2 bytes |
never@2118 | 1866 | __ andcc(count, 1<<(shift-1), G0); // in delay slot of branches |
never@2118 | 1867 | __ BIND(L_fill_2_bytes); |
never@2118 | 1868 | __ brx(Assembler::zero, false, Assembler::pt, L_fill_byte); |
never@2118 | 1869 | __ delayed()->andcc(count, 1, count); |
never@2118 | 1870 | __ sth(value, to, 0); |
never@2118 | 1871 | if (t == T_BYTE) { |
never@2118 | 1872 | __ inc(to, 2); |
never@2118 | 1873 | // fill trailing byte |
never@2118 | 1874 | __ andcc(count, 1, count); // in delay slot of branches |
never@2118 | 1875 | __ BIND(L_fill_byte); |
never@2118 | 1876 | __ brx(Assembler::zero, false, Assembler::pt, L_exit); |
never@2118 | 1877 | __ delayed()->nop(); |
never@2118 | 1878 | __ stb(value, to, 0); |
never@2118 | 1879 | } else { |
never@2118 | 1880 | __ BIND(L_fill_byte); |
never@2118 | 1881 | } |
never@2118 | 1882 | } else { |
never@2118 | 1883 | __ BIND(L_fill_2_bytes); |
never@2118 | 1884 | } |
never@2118 | 1885 | __ BIND(L_exit); |
never@2118 | 1886 | __ retl(); |
never@2149 | 1887 | __ delayed()->nop(); |
never@2149 | 1888 | |
never@2149 | 1889 | // Handle copies less than 8 bytes. Int is handled elsewhere. |
never@2149 | 1890 | if (t == T_BYTE) { |
never@2149 | 1891 | __ BIND(L_fill_elements); |
never@2149 | 1892 | Label L_fill_2, L_fill_4; |
never@2149 | 1893 | // in delay slot __ andcc(count, 1, G0); |
never@2149 | 1894 | __ brx(Assembler::zero, false, Assembler::pt, L_fill_2); |
never@2149 | 1895 | __ delayed()->andcc(count, 2, G0); |
never@2149 | 1896 | __ stb(value, to, 0); |
never@2149 | 1897 | __ inc(to, 1); |
never@2149 | 1898 | __ BIND(L_fill_2); |
never@2149 | 1899 | __ brx(Assembler::zero, false, Assembler::pt, L_fill_4); |
never@2149 | 1900 | __ delayed()->andcc(count, 4, G0); |
never@2149 | 1901 | __ stb(value, to, 0); |
never@2149 | 1902 | __ stb(value, to, 1); |
never@2149 | 1903 | __ inc(to, 2); |
never@2149 | 1904 | __ BIND(L_fill_4); |
never@2149 | 1905 | __ brx(Assembler::zero, false, Assembler::pt, L_exit); |
never@2149 | 1906 | __ delayed()->nop(); |
never@2149 | 1907 | __ stb(value, to, 0); |
never@2149 | 1908 | __ stb(value, to, 1); |
never@2149 | 1909 | __ stb(value, to, 2); |
never@2149 | 1910 | __ retl(); |
never@2149 | 1911 | __ delayed()->stb(value, to, 3); |
never@2149 | 1912 | } |
never@2149 | 1913 | |
never@2149 | 1914 | if (t == T_SHORT) { |
never@2149 | 1915 | Label L_fill_2; |
never@2149 | 1916 | __ BIND(L_fill_elements); |
never@2149 | 1917 | // in delay slot __ andcc(count, 1, G0); |
never@2149 | 1918 | __ brx(Assembler::zero, false, Assembler::pt, L_fill_2); |
never@2149 | 1919 | __ delayed()->andcc(count, 2, G0); |
never@2149 | 1920 | __ sth(value, to, 0); |
never@2149 | 1921 | __ inc(to, 2); |
never@2149 | 1922 | __ BIND(L_fill_2); |
never@2149 | 1923 | __ brx(Assembler::zero, false, Assembler::pt, L_exit); |
never@2149 | 1924 | __ delayed()->nop(); |
never@2149 | 1925 | __ sth(value, to, 0); |
never@2149 | 1926 | __ retl(); |
never@2149 | 1927 | __ delayed()->sth(value, to, 2); |
never@2149 | 1928 | } |
never@2118 | 1929 | return start; |
never@2118 | 1930 | } |
never@2118 | 1931 | |
never@2118 | 1932 | // |
duke@435 | 1933 | // Generate stub for conjoint short copy. If "aligned" is true, the |
duke@435 | 1934 | // "from" and "to" addresses are assumed to be heapword aligned. |
duke@435 | 1935 | // |
duke@435 | 1936 | // Arguments for generated stub: |
duke@435 | 1937 | // from: O0 |
duke@435 | 1938 | // to: O1 |
duke@435 | 1939 | // count: O2 treated as signed |
duke@435 | 1940 | // |
iveresov@2595 | 1941 | address generate_conjoint_short_copy(bool aligned, address nooverlap_target, |
iveresov@2595 | 1942 | address *entry, const char *name) { |
duke@435 | 1943 | // Do reverse copy. |
duke@435 | 1944 | |
duke@435 | 1945 | __ align(CodeEntryAlignment); |
duke@435 | 1946 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 1947 | address start = __ pc(); |
duke@435 | 1948 | |
duke@435 | 1949 | Label L_skip_alignment, L_skip_alignment2, L_aligned_copy; |
duke@435 | 1950 | Label L_copy_2_bytes, L_copy_2_bytes_loop, L_exit; |
duke@435 | 1951 | |
duke@435 | 1952 | const Register from = O0; // source array address |
duke@435 | 1953 | const Register to = O1; // destination array address |
duke@435 | 1954 | const Register count = O2; // elements count |
duke@435 | 1955 | const Register end_from = from; // source array end address |
duke@435 | 1956 | const Register end_to = to; // destination array end address |
duke@435 | 1957 | |
duke@435 | 1958 | const Register byte_count = O3; // bytes count to copy |
duke@435 | 1959 | |
duke@435 | 1960 | assert_clean_int(count, O3); // Make sure 'count' is clean int. |
duke@435 | 1961 | |
iveresov@2595 | 1962 | if (entry != NULL) { |
iveresov@2595 | 1963 | *entry = __ pc(); |
iveresov@2595 | 1964 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 1965 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 1966 | } |
duke@435 | 1967 | |
duke@435 | 1968 | array_overlap_test(nooverlap_target, 1); |
duke@435 | 1969 | |
duke@435 | 1970 | __ sllx(count, LogBytesPerShort, byte_count); |
duke@435 | 1971 | __ add(to, byte_count, end_to); // offset after last copied element |
duke@435 | 1972 | |
duke@435 | 1973 | // for short arrays, just do single element copy |
duke@435 | 1974 | __ cmp(count, 11); // 8 + 3 (22 bytes) |
duke@435 | 1975 | __ brx(Assembler::less, false, Assembler::pn, L_copy_2_bytes); |
duke@435 | 1976 | __ delayed()->add(from, byte_count, end_from); |
duke@435 | 1977 | |
duke@435 | 1978 | { |
duke@435 | 1979 | // Align end of arrays since they could be not aligned even |
duke@435 | 1980 | // when arrays itself are aligned. |
duke@435 | 1981 | |
duke@435 | 1982 | // copy 1 element if necessary to align 'end_to' on an 4 bytes |
duke@435 | 1983 | __ andcc(end_to, 3, G0); |
duke@435 | 1984 | __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); |
duke@435 | 1985 | __ delayed()->lduh(end_from, -2, O3); |
duke@435 | 1986 | __ dec(end_from, 2); |
duke@435 | 1987 | __ dec(end_to, 2); |
duke@435 | 1988 | __ dec(count); |
duke@435 | 1989 | __ sth(O3, end_to, 0); |
duke@435 | 1990 | __ BIND(L_skip_alignment); |
duke@435 | 1991 | |
duke@435 | 1992 | // copy 2 elements to align 'end_to' on an 8 byte boundary |
duke@435 | 1993 | __ andcc(end_to, 7, G0); |
duke@435 | 1994 | __ br(Assembler::zero, false, Assembler::pn, L_skip_alignment2); |
duke@435 | 1995 | __ delayed()->lduh(end_from, -2, O3); |
duke@435 | 1996 | __ dec(count, 2); |
duke@435 | 1997 | __ lduh(end_from, -4, O4); |
duke@435 | 1998 | __ dec(end_from, 4); |
duke@435 | 1999 | __ dec(end_to, 4); |
duke@435 | 2000 | __ sth(O3, end_to, 2); |
duke@435 | 2001 | __ sth(O4, end_to, 0); |
duke@435 | 2002 | __ BIND(L_skip_alignment2); |
duke@435 | 2003 | } |
duke@435 | 2004 | #ifdef _LP64 |
duke@435 | 2005 | if (aligned) { |
duke@435 | 2006 | // Both arrays are aligned to 8-bytes in 64-bits VM. |
duke@435 | 2007 | // The 'count' is decremented in copy_16_bytes_backward_with_shift() |
duke@435 | 2008 | // in unaligned case. |
duke@435 | 2009 | __ dec(count, 8); |
duke@435 | 2010 | } else |
duke@435 | 2011 | #endif |
duke@435 | 2012 | { |
duke@435 | 2013 | // Copy with shift 16 bytes per iteration if arrays do not have |
duke@435 | 2014 | // the same alignment mod 8, otherwise jump to the next |
duke@435 | 2015 | // code for aligned copy (and substracting 8 from 'count' before jump). |
duke@435 | 2016 | // The compare above (count >= 11) guarantes 'count' >= 16 bytes. |
duke@435 | 2017 | // Also jump over aligned copy after the copy with shift completed. |
duke@435 | 2018 | |
duke@435 | 2019 | copy_16_bytes_backward_with_shift(end_from, end_to, count, 8, |
duke@435 | 2020 | L_aligned_copy, L_copy_2_bytes); |
duke@435 | 2021 | } |
duke@435 | 2022 | // copy 4 elements (16 bytes) at a time |
kvn@1800 | 2023 | __ align(OptoLoopAlignment); |
duke@435 | 2024 | __ BIND(L_aligned_copy); |
duke@435 | 2025 | __ dec(end_from, 16); |
duke@435 | 2026 | __ ldx(end_from, 8, O3); |
duke@435 | 2027 | __ ldx(end_from, 0, O4); |
duke@435 | 2028 | __ dec(end_to, 16); |
duke@435 | 2029 | __ deccc(count, 8); |
duke@435 | 2030 | __ stx(O3, end_to, 8); |
duke@435 | 2031 | __ brx(Assembler::greaterEqual, false, Assembler::pt, L_aligned_copy); |
duke@435 | 2032 | __ delayed()->stx(O4, end_to, 0); |
duke@435 | 2033 | __ inc(count, 8); |
duke@435 | 2034 | |
duke@435 | 2035 | // copy 1 element (2 bytes) at a time |
duke@435 | 2036 | __ BIND(L_copy_2_bytes); |
kvn@3037 | 2037 | __ cmp_and_br_short(count, 0, Assembler::equal, Assembler::pt, L_exit); |
duke@435 | 2038 | __ BIND(L_copy_2_bytes_loop); |
duke@435 | 2039 | __ dec(end_from, 2); |
duke@435 | 2040 | __ dec(end_to, 2); |
duke@435 | 2041 | __ lduh(end_from, 0, O4); |
duke@435 | 2042 | __ deccc(count); |
duke@435 | 2043 | __ brx(Assembler::greater, false, Assembler::pt, L_copy_2_bytes_loop); |
duke@435 | 2044 | __ delayed()->sth(O4, end_to, 0); |
duke@435 | 2045 | |
duke@435 | 2046 | __ BIND(L_exit); |
duke@435 | 2047 | // O3, O4 are used as temp registers |
duke@435 | 2048 | inc_counter_np(SharedRuntime::_jshort_array_copy_ctr, O3, O4); |
duke@435 | 2049 | __ retl(); |
duke@435 | 2050 | __ delayed()->mov(G0, O0); // return 0 |
duke@435 | 2051 | return start; |
duke@435 | 2052 | } |
duke@435 | 2053 | |
duke@435 | 2054 | // |
kvn@3103 | 2055 | // Helper methods for generate_disjoint_int_copy_core() |
kvn@3103 | 2056 | // |
kvn@3103 | 2057 | void copy_16_bytes_loop(Register from, Register to, Register count, int count_dec, |
kvn@3103 | 2058 | Label& L_loop, bool use_prefetch, bool use_bis) { |
kvn@3103 | 2059 | |
kvn@3103 | 2060 | __ align(OptoLoopAlignment); |
kvn@3103 | 2061 | __ BIND(L_loop); |
kvn@3103 | 2062 | if (use_prefetch) { |
kvn@3103 | 2063 | if (ArraycopySrcPrefetchDistance > 0) { |
kvn@3103 | 2064 | __ prefetch(from, ArraycopySrcPrefetchDistance, Assembler::severalReads); |
kvn@3103 | 2065 | } |
kvn@3103 | 2066 | if (ArraycopyDstPrefetchDistance > 0) { |
kvn@3103 | 2067 | __ prefetch(to, ArraycopyDstPrefetchDistance, Assembler::severalWritesAndPossiblyReads); |
kvn@3103 | 2068 | } |
kvn@3103 | 2069 | } |
kvn@3103 | 2070 | __ ldx(from, 4, O4); |
kvn@3103 | 2071 | __ ldx(from, 12, G4); |
kvn@3103 | 2072 | __ inc(to, 16); |
kvn@3103 | 2073 | __ inc(from, 16); |
kvn@3103 | 2074 | __ deccc(count, 4); // Can we do next iteration after this one? |
kvn@3103 | 2075 | |
kvn@3103 | 2076 | __ srlx(O4, 32, G3); |
kvn@3103 | 2077 | __ bset(G3, O3); |
kvn@3103 | 2078 | __ sllx(O4, 32, O4); |
kvn@3103 | 2079 | __ srlx(G4, 32, G3); |
kvn@3103 | 2080 | __ bset(G3, O4); |
kvn@3103 | 2081 | if (use_bis) { |
kvn@3103 | 2082 | __ stxa(O3, to, -16); |
kvn@3103 | 2083 | __ stxa(O4, to, -8); |
kvn@3103 | 2084 | } else { |
kvn@3103 | 2085 | __ stx(O3, to, -16); |
kvn@3103 | 2086 | __ stx(O4, to, -8); |
kvn@3103 | 2087 | } |
kvn@3103 | 2088 | __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop); |
kvn@3103 | 2089 | __ delayed()->sllx(G4, 32, O3); |
kvn@3103 | 2090 | |
kvn@3103 | 2091 | } |
kvn@3103 | 2092 | |
kvn@3103 | 2093 | // |
duke@435 | 2094 | // Generate core code for disjoint int copy (and oop copy on 32-bit). |
duke@435 | 2095 | // If "aligned" is true, the "from" and "to" addresses are assumed |
duke@435 | 2096 | // to be heapword aligned. |
duke@435 | 2097 | // |
duke@435 | 2098 | // Arguments: |
duke@435 | 2099 | // from: O0 |
duke@435 | 2100 | // to: O1 |
duke@435 | 2101 | // count: O2 treated as signed |
duke@435 | 2102 | // |
duke@435 | 2103 | void generate_disjoint_int_copy_core(bool aligned) { |
duke@435 | 2104 | |
duke@435 | 2105 | Label L_skip_alignment, L_aligned_copy; |
kvn@3103 | 2106 | Label L_copy_4_bytes, L_copy_4_bytes_loop, L_exit; |
duke@435 | 2107 | |
duke@435 | 2108 | const Register from = O0; // source array address |
duke@435 | 2109 | const Register to = O1; // destination array address |
duke@435 | 2110 | const Register count = O2; // elements count |
duke@435 | 2111 | const Register offset = O5; // offset from start of arrays |
duke@435 | 2112 | // O3, O4, G3, G4 are used as temp registers |
duke@435 | 2113 | |
duke@435 | 2114 | // 'aligned' == true when it is known statically during compilation |
duke@435 | 2115 | // of this arraycopy call site that both 'from' and 'to' addresses |
duke@435 | 2116 | // are HeapWordSize aligned (see LibraryCallKit::basictype2arraycopy()). |
duke@435 | 2117 | // |
duke@435 | 2118 | // Aligned arrays have 4 bytes alignment in 32-bits VM |
duke@435 | 2119 | // and 8 bytes - in 64-bits VM. |
duke@435 | 2120 | // |
duke@435 | 2121 | #ifdef _LP64 |
duke@435 | 2122 | if (!aligned) |
duke@435 | 2123 | #endif |
duke@435 | 2124 | { |
duke@435 | 2125 | // The next check could be put under 'ifndef' since the code in |
duke@435 | 2126 | // generate_disjoint_long_copy_core() has own checks and set 'offset'. |
duke@435 | 2127 | |
duke@435 | 2128 | // for short arrays, just do single element copy |
duke@435 | 2129 | __ cmp(count, 5); // 4 + 1 (20 bytes) |
duke@435 | 2130 | __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_4_bytes); |
duke@435 | 2131 | __ delayed()->mov(G0, offset); |
duke@435 | 2132 | |
duke@435 | 2133 | // copy 1 element to align 'to' on an 8 byte boundary |
duke@435 | 2134 | __ andcc(to, 7, G0); |
duke@435 | 2135 | __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); |
duke@435 | 2136 | __ delayed()->ld(from, 0, O3); |
duke@435 | 2137 | __ inc(from, 4); |
duke@435 | 2138 | __ inc(to, 4); |
duke@435 | 2139 | __ dec(count); |
duke@435 | 2140 | __ st(O3, to, -4); |
duke@435 | 2141 | __ BIND(L_skip_alignment); |
duke@435 | 2142 | |
duke@435 | 2143 | // if arrays have same alignment mod 8, do 4 elements copy |
duke@435 | 2144 | __ andcc(from, 7, G0); |
duke@435 | 2145 | __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy); |
duke@435 | 2146 | __ delayed()->ld(from, 0, O3); |
duke@435 | 2147 | |
duke@435 | 2148 | // |
duke@435 | 2149 | // Load 2 aligned 8-bytes chunks and use one from previous iteration |
duke@435 | 2150 | // to form 2 aligned 8-bytes chunks to store. |
duke@435 | 2151 | // |
duke@435 | 2152 | // copy_16_bytes_forward_with_shift() is not used here since this |
duke@435 | 2153 | // code is more optimal. |
duke@435 | 2154 | |
duke@435 | 2155 | // copy with shift 4 elements (16 bytes) at a time |
duke@435 | 2156 | __ dec(count, 4); // The cmp at the beginning guaranty count >= 4 |
kvn@3103 | 2157 | __ sllx(O3, 32, O3); |
kvn@3103 | 2158 | |
kvn@3103 | 2159 | disjoint_copy_core(from, to, count, 2, 16, copy_16_bytes_loop); |
duke@435 | 2160 | |
duke@435 | 2161 | __ br(Assembler::always, false, Assembler::pt, L_copy_4_bytes); |
duke@435 | 2162 | __ delayed()->inc(count, 4); // restore 'count' |
duke@435 | 2163 | |
duke@435 | 2164 | __ BIND(L_aligned_copy); |
kvn@3103 | 2165 | } // !aligned |
kvn@3103 | 2166 | |
duke@435 | 2167 | // copy 4 elements (16 bytes) at a time |
duke@435 | 2168 | __ and3(count, 1, G4); // Save |
duke@435 | 2169 | __ srl(count, 1, count); |
duke@435 | 2170 | generate_disjoint_long_copy_core(aligned); |
duke@435 | 2171 | __ mov(G4, count); // Restore |
duke@435 | 2172 | |
duke@435 | 2173 | // copy 1 element at a time |
duke@435 | 2174 | __ BIND(L_copy_4_bytes); |
kvn@3037 | 2175 | __ cmp_and_br_short(count, 0, Assembler::equal, Assembler::pt, L_exit); |
duke@435 | 2176 | __ BIND(L_copy_4_bytes_loop); |
duke@435 | 2177 | __ ld(from, offset, O3); |
duke@435 | 2178 | __ deccc(count); |
duke@435 | 2179 | __ st(O3, to, offset); |
duke@435 | 2180 | __ brx(Assembler::notZero, false, Assembler::pt, L_copy_4_bytes_loop); |
duke@435 | 2181 | __ delayed()->inc(offset, 4); |
duke@435 | 2182 | __ BIND(L_exit); |
duke@435 | 2183 | } |
duke@435 | 2184 | |
duke@435 | 2185 | // |
duke@435 | 2186 | // Generate stub for disjoint int copy. If "aligned" is true, the |
duke@435 | 2187 | // "from" and "to" addresses are assumed to be heapword aligned. |
duke@435 | 2188 | // |
duke@435 | 2189 | // Arguments for generated stub: |
duke@435 | 2190 | // from: O0 |
duke@435 | 2191 | // to: O1 |
duke@435 | 2192 | // count: O2 treated as signed |
duke@435 | 2193 | // |
iveresov@2595 | 2194 | address generate_disjoint_int_copy(bool aligned, address *entry, const char *name) { |
duke@435 | 2195 | __ align(CodeEntryAlignment); |
duke@435 | 2196 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2197 | address start = __ pc(); |
duke@435 | 2198 | |
duke@435 | 2199 | const Register count = O2; |
duke@435 | 2200 | assert_clean_int(count, O3); // Make sure 'count' is clean int. |
duke@435 | 2201 | |
iveresov@2595 | 2202 | if (entry != NULL) { |
iveresov@2595 | 2203 | *entry = __ pc(); |
iveresov@2595 | 2204 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 2205 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 2206 | } |
duke@435 | 2207 | |
duke@435 | 2208 | generate_disjoint_int_copy_core(aligned); |
duke@435 | 2209 | |
duke@435 | 2210 | // O3, O4 are used as temp registers |
duke@435 | 2211 | inc_counter_np(SharedRuntime::_jint_array_copy_ctr, O3, O4); |
duke@435 | 2212 | __ retl(); |
duke@435 | 2213 | __ delayed()->mov(G0, O0); // return 0 |
duke@435 | 2214 | return start; |
duke@435 | 2215 | } |
duke@435 | 2216 | |
duke@435 | 2217 | // |
duke@435 | 2218 | // Generate core code for conjoint int copy (and oop copy on 32-bit). |
duke@435 | 2219 | // If "aligned" is true, the "from" and "to" addresses are assumed |
duke@435 | 2220 | // to be heapword aligned. |
duke@435 | 2221 | // |
duke@435 | 2222 | // Arguments: |
duke@435 | 2223 | // from: O0 |
duke@435 | 2224 | // to: O1 |
duke@435 | 2225 | // count: O2 treated as signed |
duke@435 | 2226 | // |
duke@435 | 2227 | void generate_conjoint_int_copy_core(bool aligned) { |
duke@435 | 2228 | // Do reverse copy. |
duke@435 | 2229 | |
duke@435 | 2230 | Label L_skip_alignment, L_aligned_copy; |
duke@435 | 2231 | Label L_copy_16_bytes, L_copy_4_bytes, L_copy_4_bytes_loop, L_exit; |
duke@435 | 2232 | |
duke@435 | 2233 | const Register from = O0; // source array address |
duke@435 | 2234 | const Register to = O1; // destination array address |
duke@435 | 2235 | const Register count = O2; // elements count |
duke@435 | 2236 | const Register end_from = from; // source array end address |
duke@435 | 2237 | const Register end_to = to; // destination array end address |
duke@435 | 2238 | // O3, O4, O5, G3 are used as temp registers |
duke@435 | 2239 | |
duke@435 | 2240 | const Register byte_count = O3; // bytes count to copy |
duke@435 | 2241 | |
duke@435 | 2242 | __ sllx(count, LogBytesPerInt, byte_count); |
duke@435 | 2243 | __ add(to, byte_count, end_to); // offset after last copied element |
duke@435 | 2244 | |
duke@435 | 2245 | __ cmp(count, 5); // for short arrays, just do single element copy |
duke@435 | 2246 | __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_4_bytes); |
duke@435 | 2247 | __ delayed()->add(from, byte_count, end_from); |
duke@435 | 2248 | |
duke@435 | 2249 | // copy 1 element to align 'to' on an 8 byte boundary |
duke@435 | 2250 | __ andcc(end_to, 7, G0); |
duke@435 | 2251 | __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); |
duke@435 | 2252 | __ delayed()->nop(); |
duke@435 | 2253 | __ dec(count); |
duke@435 | 2254 | __ dec(end_from, 4); |
duke@435 | 2255 | __ dec(end_to, 4); |
duke@435 | 2256 | __ ld(end_from, 0, O4); |
duke@435 | 2257 | __ st(O4, end_to, 0); |
duke@435 | 2258 | __ BIND(L_skip_alignment); |
duke@435 | 2259 | |
duke@435 | 2260 | // Check if 'end_from' and 'end_to' has the same alignment. |
duke@435 | 2261 | __ andcc(end_from, 7, G0); |
duke@435 | 2262 | __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy); |
duke@435 | 2263 | __ delayed()->dec(count, 4); // The cmp at the start guaranty cnt >= 4 |
duke@435 | 2264 | |
duke@435 | 2265 | // copy with shift 4 elements (16 bytes) at a time |
duke@435 | 2266 | // |
duke@435 | 2267 | // Load 2 aligned 8-bytes chunks and use one from previous iteration |
duke@435 | 2268 | // to form 2 aligned 8-bytes chunks to store. |
duke@435 | 2269 | // |
duke@435 | 2270 | __ ldx(end_from, -4, O3); |
kvn@1800 | 2271 | __ align(OptoLoopAlignment); |
duke@435 | 2272 | __ BIND(L_copy_16_bytes); |
duke@435 | 2273 | __ ldx(end_from, -12, O4); |
duke@435 | 2274 | __ deccc(count, 4); |
duke@435 | 2275 | __ ldx(end_from, -20, O5); |
duke@435 | 2276 | __ dec(end_to, 16); |
duke@435 | 2277 | __ dec(end_from, 16); |
duke@435 | 2278 | __ srlx(O3, 32, O3); |
duke@435 | 2279 | __ sllx(O4, 32, G3); |
duke@435 | 2280 | __ bset(G3, O3); |
duke@435 | 2281 | __ stx(O3, end_to, 8); |
duke@435 | 2282 | __ srlx(O4, 32, O4); |
duke@435 | 2283 | __ sllx(O5, 32, G3); |
duke@435 | 2284 | __ bset(O4, G3); |
duke@435 | 2285 | __ stx(G3, end_to, 0); |
duke@435 | 2286 | __ brx(Assembler::greaterEqual, false, Assembler::pt, L_copy_16_bytes); |
duke@435 | 2287 | __ delayed()->mov(O5, O3); |
duke@435 | 2288 | |
duke@435 | 2289 | __ br(Assembler::always, false, Assembler::pt, L_copy_4_bytes); |
duke@435 | 2290 | __ delayed()->inc(count, 4); |
duke@435 | 2291 | |
duke@435 | 2292 | // copy 4 elements (16 bytes) at a time |
kvn@1800 | 2293 | __ align(OptoLoopAlignment); |
duke@435 | 2294 | __ BIND(L_aligned_copy); |
duke@435 | 2295 | __ dec(end_from, 16); |
duke@435 | 2296 | __ ldx(end_from, 8, O3); |
duke@435 | 2297 | __ ldx(end_from, 0, O4); |
duke@435 | 2298 | __ dec(end_to, 16); |
duke@435 | 2299 | __ deccc(count, 4); |
duke@435 | 2300 | __ stx(O3, end_to, 8); |
duke@435 | 2301 | __ brx(Assembler::greaterEqual, false, Assembler::pt, L_aligned_copy); |
duke@435 | 2302 | __ delayed()->stx(O4, end_to, 0); |
duke@435 | 2303 | __ inc(count, 4); |
duke@435 | 2304 | |
duke@435 | 2305 | // copy 1 element (4 bytes) at a time |
duke@435 | 2306 | __ BIND(L_copy_4_bytes); |
kvn@3037 | 2307 | __ cmp_and_br_short(count, 0, Assembler::equal, Assembler::pt, L_exit); |
duke@435 | 2308 | __ BIND(L_copy_4_bytes_loop); |
duke@435 | 2309 | __ dec(end_from, 4); |
duke@435 | 2310 | __ dec(end_to, 4); |
duke@435 | 2311 | __ ld(end_from, 0, O4); |
duke@435 | 2312 | __ deccc(count); |
duke@435 | 2313 | __ brx(Assembler::greater, false, Assembler::pt, L_copy_4_bytes_loop); |
duke@435 | 2314 | __ delayed()->st(O4, end_to, 0); |
duke@435 | 2315 | __ BIND(L_exit); |
duke@435 | 2316 | } |
duke@435 | 2317 | |
duke@435 | 2318 | // |
duke@435 | 2319 | // Generate stub for conjoint int copy. If "aligned" is true, the |
duke@435 | 2320 | // "from" and "to" addresses are assumed to be heapword aligned. |
duke@435 | 2321 | // |
duke@435 | 2322 | // Arguments for generated stub: |
duke@435 | 2323 | // from: O0 |
duke@435 | 2324 | // to: O1 |
duke@435 | 2325 | // count: O2 treated as signed |
duke@435 | 2326 | // |
iveresov@2595 | 2327 | address generate_conjoint_int_copy(bool aligned, address nooverlap_target, |
iveresov@2595 | 2328 | address *entry, const char *name) { |
duke@435 | 2329 | __ align(CodeEntryAlignment); |
duke@435 | 2330 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2331 | address start = __ pc(); |
duke@435 | 2332 | |
duke@435 | 2333 | assert_clean_int(O2, O3); // Make sure 'count' is clean int. |
duke@435 | 2334 | |
iveresov@2595 | 2335 | if (entry != NULL) { |
iveresov@2595 | 2336 | *entry = __ pc(); |
iveresov@2595 | 2337 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 2338 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 2339 | } |
duke@435 | 2340 | |
duke@435 | 2341 | array_overlap_test(nooverlap_target, 2); |
duke@435 | 2342 | |
duke@435 | 2343 | generate_conjoint_int_copy_core(aligned); |
duke@435 | 2344 | |
duke@435 | 2345 | // O3, O4 are used as temp registers |
duke@435 | 2346 | inc_counter_np(SharedRuntime::_jint_array_copy_ctr, O3, O4); |
duke@435 | 2347 | __ retl(); |
duke@435 | 2348 | __ delayed()->mov(G0, O0); // return 0 |
duke@435 | 2349 | return start; |
duke@435 | 2350 | } |
duke@435 | 2351 | |
duke@435 | 2352 | // |
kvn@3103 | 2353 | // Helper methods for generate_disjoint_long_copy_core() |
kvn@3103 | 2354 | // |
kvn@3103 | 2355 | void copy_64_bytes_loop(Register from, Register to, Register count, int count_dec, |
kvn@3103 | 2356 | Label& L_loop, bool use_prefetch, bool use_bis) { |
kvn@3103 | 2357 | __ align(OptoLoopAlignment); |
kvn@3103 | 2358 | __ BIND(L_loop); |
kvn@3103 | 2359 | for (int off = 0; off < 64; off += 16) { |
kvn@3103 | 2360 | if (use_prefetch && (off & 31) == 0) { |
kvn@3103 | 2361 | if (ArraycopySrcPrefetchDistance > 0) { |
kvn@3157 | 2362 | __ prefetch(from, ArraycopySrcPrefetchDistance+off, Assembler::severalReads); |
kvn@3103 | 2363 | } |
kvn@3103 | 2364 | if (ArraycopyDstPrefetchDistance > 0) { |
kvn@3157 | 2365 | __ prefetch(to, ArraycopyDstPrefetchDistance+off, Assembler::severalWritesAndPossiblyReads); |
kvn@3103 | 2366 | } |
kvn@3103 | 2367 | } |
kvn@3103 | 2368 | __ ldx(from, off+0, O4); |
kvn@3103 | 2369 | __ ldx(from, off+8, O5); |
kvn@3103 | 2370 | if (use_bis) { |
kvn@3103 | 2371 | __ stxa(O4, to, off+0); |
kvn@3103 | 2372 | __ stxa(O5, to, off+8); |
kvn@3103 | 2373 | } else { |
kvn@3103 | 2374 | __ stx(O4, to, off+0); |
kvn@3103 | 2375 | __ stx(O5, to, off+8); |
kvn@3103 | 2376 | } |
kvn@3103 | 2377 | } |
kvn@3103 | 2378 | __ deccc(count, 8); |
kvn@3103 | 2379 | __ inc(from, 64); |
kvn@3103 | 2380 | __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop); |
kvn@3103 | 2381 | __ delayed()->inc(to, 64); |
kvn@3103 | 2382 | } |
kvn@3103 | 2383 | |
kvn@3103 | 2384 | // |
duke@435 | 2385 | // Generate core code for disjoint long copy (and oop copy on 64-bit). |
duke@435 | 2386 | // "aligned" is ignored, because we must make the stronger |
duke@435 | 2387 | // assumption that both addresses are always 64-bit aligned. |
duke@435 | 2388 | // |
duke@435 | 2389 | // Arguments: |
duke@435 | 2390 | // from: O0 |
duke@435 | 2391 | // to: O1 |
duke@435 | 2392 | // count: O2 treated as signed |
duke@435 | 2393 | // |
kvn@1799 | 2394 | // count -= 2; |
kvn@1799 | 2395 | // if ( count >= 0 ) { // >= 2 elements |
kvn@1799 | 2396 | // if ( count > 6) { // >= 8 elements |
kvn@1799 | 2397 | // count -= 6; // original count - 8 |
kvn@1799 | 2398 | // do { |
kvn@1799 | 2399 | // copy_8_elements; |
kvn@1799 | 2400 | // count -= 8; |
kvn@1799 | 2401 | // } while ( count >= 0 ); |
kvn@1799 | 2402 | // count += 6; |
kvn@1799 | 2403 | // } |
kvn@1799 | 2404 | // if ( count >= 0 ) { // >= 2 elements |
kvn@1799 | 2405 | // do { |
kvn@1799 | 2406 | // copy_2_elements; |
kvn@1799 | 2407 | // } while ( (count=count-2) >= 0 ); |
kvn@1799 | 2408 | // } |
kvn@1799 | 2409 | // } |
kvn@1799 | 2410 | // count += 2; |
kvn@1799 | 2411 | // if ( count != 0 ) { // 1 element left |
kvn@1799 | 2412 | // copy_1_element; |
kvn@1799 | 2413 | // } |
kvn@1799 | 2414 | // |
duke@435 | 2415 | void generate_disjoint_long_copy_core(bool aligned) { |
duke@435 | 2416 | Label L_copy_8_bytes, L_copy_16_bytes, L_exit; |
duke@435 | 2417 | const Register from = O0; // source array address |
duke@435 | 2418 | const Register to = O1; // destination array address |
duke@435 | 2419 | const Register count = O2; // elements count |
duke@435 | 2420 | const Register offset0 = O4; // element offset |
duke@435 | 2421 | const Register offset8 = O5; // next element offset |
duke@435 | 2422 | |
kvn@3103 | 2423 | __ deccc(count, 2); |
kvn@3103 | 2424 | __ mov(G0, offset0); // offset from start of arrays (0) |
kvn@3103 | 2425 | __ brx(Assembler::negative, false, Assembler::pn, L_copy_8_bytes ); |
kvn@3103 | 2426 | __ delayed()->add(offset0, 8, offset8); |
kvn@1799 | 2427 | |
kvn@1799 | 2428 | // Copy by 64 bytes chunks |
kvn@3103 | 2429 | |
kvn@1799 | 2430 | const Register from64 = O3; // source address |
kvn@1799 | 2431 | const Register to64 = G3; // destination address |
kvn@3103 | 2432 | __ subcc(count, 6, O3); |
kvn@3103 | 2433 | __ brx(Assembler::negative, false, Assembler::pt, L_copy_16_bytes ); |
kvn@3103 | 2434 | __ delayed()->mov(to, to64); |
kvn@3103 | 2435 | // Now we can use O4(offset0), O5(offset8) as temps |
kvn@3103 | 2436 | __ mov(O3, count); |
kvn@3103 | 2437 | // count >= 0 (original count - 8) |
kvn@3103 | 2438 | __ mov(from, from64); |
kvn@3103 | 2439 | |
kvn@3103 | 2440 | disjoint_copy_core(from64, to64, count, 3, 64, copy_64_bytes_loop); |
kvn@1799 | 2441 | |
kvn@1799 | 2442 | // Restore O4(offset0), O5(offset8) |
kvn@1799 | 2443 | __ sub(from64, from, offset0); |
kvn@3103 | 2444 | __ inccc(count, 6); // restore count |
kvn@1799 | 2445 | __ brx(Assembler::negative, false, Assembler::pn, L_copy_8_bytes ); |
kvn@1799 | 2446 | __ delayed()->add(offset0, 8, offset8); |
kvn@1799 | 2447 | |
kvn@1799 | 2448 | // Copy by 16 bytes chunks |
kvn@1800 | 2449 | __ align(OptoLoopAlignment); |
duke@435 | 2450 | __ BIND(L_copy_16_bytes); |
duke@435 | 2451 | __ ldx(from, offset0, O3); |
duke@435 | 2452 | __ ldx(from, offset8, G3); |
duke@435 | 2453 | __ deccc(count, 2); |
duke@435 | 2454 | __ stx(O3, to, offset0); |
duke@435 | 2455 | __ inc(offset0, 16); |
duke@435 | 2456 | __ stx(G3, to, offset8); |
duke@435 | 2457 | __ brx(Assembler::greaterEqual, false, Assembler::pt, L_copy_16_bytes); |
duke@435 | 2458 | __ delayed()->inc(offset8, 16); |
duke@435 | 2459 | |
kvn@1799 | 2460 | // Copy last 8 bytes |
duke@435 | 2461 | __ BIND(L_copy_8_bytes); |
duke@435 | 2462 | __ inccc(count, 2); |
duke@435 | 2463 | __ brx(Assembler::zero, true, Assembler::pn, L_exit ); |
duke@435 | 2464 | __ delayed()->mov(offset0, offset8); // Set O5 used by other stubs |
duke@435 | 2465 | __ ldx(from, offset0, O3); |
duke@435 | 2466 | __ stx(O3, to, offset0); |
duke@435 | 2467 | __ BIND(L_exit); |
duke@435 | 2468 | } |
duke@435 | 2469 | |
duke@435 | 2470 | // |
duke@435 | 2471 | // Generate stub for disjoint long copy. |
duke@435 | 2472 | // "aligned" is ignored, because we must make the stronger |
duke@435 | 2473 | // assumption that both addresses are always 64-bit aligned. |
duke@435 | 2474 | // |
duke@435 | 2475 | // Arguments for generated stub: |
duke@435 | 2476 | // from: O0 |
duke@435 | 2477 | // to: O1 |
duke@435 | 2478 | // count: O2 treated as signed |
duke@435 | 2479 | // |
iveresov@2595 | 2480 | address generate_disjoint_long_copy(bool aligned, address *entry, const char *name) { |
duke@435 | 2481 | __ align(CodeEntryAlignment); |
duke@435 | 2482 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2483 | address start = __ pc(); |
duke@435 | 2484 | |
duke@435 | 2485 | assert_clean_int(O2, O3); // Make sure 'count' is clean int. |
duke@435 | 2486 | |
iveresov@2595 | 2487 | if (entry != NULL) { |
iveresov@2595 | 2488 | *entry = __ pc(); |
iveresov@2595 | 2489 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 2490 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 2491 | } |
duke@435 | 2492 | |
duke@435 | 2493 | generate_disjoint_long_copy_core(aligned); |
duke@435 | 2494 | |
duke@435 | 2495 | // O3, O4 are used as temp registers |
duke@435 | 2496 | inc_counter_np(SharedRuntime::_jlong_array_copy_ctr, O3, O4); |
duke@435 | 2497 | __ retl(); |
duke@435 | 2498 | __ delayed()->mov(G0, O0); // return 0 |
duke@435 | 2499 | return start; |
duke@435 | 2500 | } |
duke@435 | 2501 | |
duke@435 | 2502 | // |
duke@435 | 2503 | // Generate core code for conjoint long copy (and oop copy on 64-bit). |
duke@435 | 2504 | // "aligned" is ignored, because we must make the stronger |
duke@435 | 2505 | // assumption that both addresses are always 64-bit aligned. |
duke@435 | 2506 | // |
duke@435 | 2507 | // Arguments: |
duke@435 | 2508 | // from: O0 |
duke@435 | 2509 | // to: O1 |
duke@435 | 2510 | // count: O2 treated as signed |
duke@435 | 2511 | // |
duke@435 | 2512 | void generate_conjoint_long_copy_core(bool aligned) { |
duke@435 | 2513 | // Do reverse copy. |
duke@435 | 2514 | Label L_copy_8_bytes, L_copy_16_bytes, L_exit; |
duke@435 | 2515 | const Register from = O0; // source array address |
duke@435 | 2516 | const Register to = O1; // destination array address |
duke@435 | 2517 | const Register count = O2; // elements count |
duke@435 | 2518 | const Register offset8 = O4; // element offset |
duke@435 | 2519 | const Register offset0 = O5; // previous element offset |
duke@435 | 2520 | |
duke@435 | 2521 | __ subcc(count, 1, count); |
duke@435 | 2522 | __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_8_bytes ); |
duke@435 | 2523 | __ delayed()->sllx(count, LogBytesPerLong, offset8); |
duke@435 | 2524 | __ sub(offset8, 8, offset0); |
kvn@1800 | 2525 | __ align(OptoLoopAlignment); |
duke@435 | 2526 | __ BIND(L_copy_16_bytes); |
duke@435 | 2527 | __ ldx(from, offset8, O2); |
duke@435 | 2528 | __ ldx(from, offset0, O3); |
duke@435 | 2529 | __ stx(O2, to, offset8); |
duke@435 | 2530 | __ deccc(offset8, 16); // use offset8 as counter |
duke@435 | 2531 | __ stx(O3, to, offset0); |
duke@435 | 2532 | __ brx(Assembler::greater, false, Assembler::pt, L_copy_16_bytes); |
duke@435 | 2533 | __ delayed()->dec(offset0, 16); |
duke@435 | 2534 | |
duke@435 | 2535 | __ BIND(L_copy_8_bytes); |
duke@435 | 2536 | __ brx(Assembler::negative, false, Assembler::pn, L_exit ); |
duke@435 | 2537 | __ delayed()->nop(); |
duke@435 | 2538 | __ ldx(from, 0, O3); |
duke@435 | 2539 | __ stx(O3, to, 0); |
duke@435 | 2540 | __ BIND(L_exit); |
duke@435 | 2541 | } |
duke@435 | 2542 | |
duke@435 | 2543 | // Generate stub for conjoint long copy. |
duke@435 | 2544 | // "aligned" is ignored, because we must make the stronger |
duke@435 | 2545 | // assumption that both addresses are always 64-bit aligned. |
duke@435 | 2546 | // |
duke@435 | 2547 | // Arguments for generated stub: |
duke@435 | 2548 | // from: O0 |
duke@435 | 2549 | // to: O1 |
duke@435 | 2550 | // count: O2 treated as signed |
duke@435 | 2551 | // |
iveresov@2595 | 2552 | address generate_conjoint_long_copy(bool aligned, address nooverlap_target, |
iveresov@2595 | 2553 | address *entry, const char *name) { |
duke@435 | 2554 | __ align(CodeEntryAlignment); |
duke@435 | 2555 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2556 | address start = __ pc(); |
duke@435 | 2557 | |
iveresov@2606 | 2558 | assert(aligned, "Should always be aligned"); |
duke@435 | 2559 | |
duke@435 | 2560 | assert_clean_int(O2, O3); // Make sure 'count' is clean int. |
duke@435 | 2561 | |
iveresov@2595 | 2562 | if (entry != NULL) { |
iveresov@2595 | 2563 | *entry = __ pc(); |
iveresov@2595 | 2564 | // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) |
iveresov@2595 | 2565 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 2566 | } |
duke@435 | 2567 | |
duke@435 | 2568 | array_overlap_test(nooverlap_target, 3); |
duke@435 | 2569 | |
duke@435 | 2570 | generate_conjoint_long_copy_core(aligned); |
duke@435 | 2571 | |
duke@435 | 2572 | // O3, O4 are used as temp registers |
duke@435 | 2573 | inc_counter_np(SharedRuntime::_jlong_array_copy_ctr, O3, O4); |
duke@435 | 2574 | __ retl(); |
duke@435 | 2575 | __ delayed()->mov(G0, O0); // return 0 |
duke@435 | 2576 | return start; |
duke@435 | 2577 | } |
duke@435 | 2578 | |
duke@435 | 2579 | // Generate stub for disjoint oop copy. If "aligned" is true, the |
duke@435 | 2580 | // "from" and "to" addresses are assumed to be heapword aligned. |
duke@435 | 2581 | // |
duke@435 | 2582 | // Arguments for generated stub: |
duke@435 | 2583 | // from: O0 |
duke@435 | 2584 | // to: O1 |
duke@435 | 2585 | // count: O2 treated as signed |
duke@435 | 2586 | // |
iveresov@2606 | 2587 | address generate_disjoint_oop_copy(bool aligned, address *entry, const char *name, |
iveresov@2606 | 2588 | bool dest_uninitialized = false) { |
duke@435 | 2589 | |
duke@435 | 2590 | const Register from = O0; // source array address |
duke@435 | 2591 | const Register to = O1; // destination array address |
duke@435 | 2592 | const Register count = O2; // elements count |
duke@435 | 2593 | |
duke@435 | 2594 | __ align(CodeEntryAlignment); |
duke@435 | 2595 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2596 | address start = __ pc(); |
duke@435 | 2597 | |
duke@435 | 2598 | assert_clean_int(count, O3); // Make sure 'count' is clean int. |
duke@435 | 2599 | |
iveresov@2595 | 2600 | if (entry != NULL) { |
iveresov@2595 | 2601 | *entry = __ pc(); |
iveresov@2595 | 2602 | // caller can pass a 64-bit byte count here |
iveresov@2595 | 2603 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 2604 | } |
duke@435 | 2605 | |
duke@435 | 2606 | // save arguments for barrier generation |
duke@435 | 2607 | __ mov(to, G1); |
duke@435 | 2608 | __ mov(count, G5); |
iveresov@2606 | 2609 | gen_write_ref_array_pre_barrier(G1, G5, dest_uninitialized); |
duke@435 | 2610 | #ifdef _LP64 |
coleenp@548 | 2611 | assert_clean_int(count, O3); // Make sure 'count' is clean int. |
coleenp@548 | 2612 | if (UseCompressedOops) { |
coleenp@548 | 2613 | generate_disjoint_int_copy_core(aligned); |
coleenp@548 | 2614 | } else { |
coleenp@548 | 2615 | generate_disjoint_long_copy_core(aligned); |
coleenp@548 | 2616 | } |
duke@435 | 2617 | #else |
duke@435 | 2618 | generate_disjoint_int_copy_core(aligned); |
duke@435 | 2619 | #endif |
duke@435 | 2620 | // O0 is used as temp register |
duke@435 | 2621 | gen_write_ref_array_post_barrier(G1, G5, O0); |
duke@435 | 2622 | |
duke@435 | 2623 | // O3, O4 are used as temp registers |
duke@435 | 2624 | inc_counter_np(SharedRuntime::_oop_array_copy_ctr, O3, O4); |
duke@435 | 2625 | __ retl(); |
duke@435 | 2626 | __ delayed()->mov(G0, O0); // return 0 |
duke@435 | 2627 | return start; |
duke@435 | 2628 | } |
duke@435 | 2629 | |
duke@435 | 2630 | // Generate stub for conjoint oop copy. If "aligned" is true, the |
duke@435 | 2631 | // "from" and "to" addresses are assumed to be heapword aligned. |
duke@435 | 2632 | // |
duke@435 | 2633 | // Arguments for generated stub: |
duke@435 | 2634 | // from: O0 |
duke@435 | 2635 | // to: O1 |
duke@435 | 2636 | // count: O2 treated as signed |
duke@435 | 2637 | // |
iveresov@2595 | 2638 | address generate_conjoint_oop_copy(bool aligned, address nooverlap_target, |
iveresov@2606 | 2639 | address *entry, const char *name, |
iveresov@2606 | 2640 | bool dest_uninitialized = false) { |
duke@435 | 2641 | |
duke@435 | 2642 | const Register from = O0; // source array address |
duke@435 | 2643 | const Register to = O1; // destination array address |
duke@435 | 2644 | const Register count = O2; // elements count |
duke@435 | 2645 | |
duke@435 | 2646 | __ align(CodeEntryAlignment); |
duke@435 | 2647 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2648 | address start = __ pc(); |
duke@435 | 2649 | |
duke@435 | 2650 | assert_clean_int(count, O3); // Make sure 'count' is clean int. |
duke@435 | 2651 | |
iveresov@2595 | 2652 | if (entry != NULL) { |
iveresov@2595 | 2653 | *entry = __ pc(); |
iveresov@2595 | 2654 | // caller can pass a 64-bit byte count here |
iveresov@2595 | 2655 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 2656 | } |
iveresov@2595 | 2657 | |
iveresov@2595 | 2658 | array_overlap_test(nooverlap_target, LogBytesPerHeapOop); |
duke@435 | 2659 | |
duke@435 | 2660 | // save arguments for barrier generation |
duke@435 | 2661 | __ mov(to, G1); |
duke@435 | 2662 | __ mov(count, G5); |
iveresov@2606 | 2663 | gen_write_ref_array_pre_barrier(G1, G5, dest_uninitialized); |
duke@435 | 2664 | |
duke@435 | 2665 | #ifdef _LP64 |
coleenp@548 | 2666 | if (UseCompressedOops) { |
coleenp@548 | 2667 | generate_conjoint_int_copy_core(aligned); |
coleenp@548 | 2668 | } else { |
coleenp@548 | 2669 | generate_conjoint_long_copy_core(aligned); |
coleenp@548 | 2670 | } |
duke@435 | 2671 | #else |
duke@435 | 2672 | generate_conjoint_int_copy_core(aligned); |
duke@435 | 2673 | #endif |
duke@435 | 2674 | |
duke@435 | 2675 | // O0 is used as temp register |
duke@435 | 2676 | gen_write_ref_array_post_barrier(G1, G5, O0); |
duke@435 | 2677 | |
duke@435 | 2678 | // O3, O4 are used as temp registers |
duke@435 | 2679 | inc_counter_np(SharedRuntime::_oop_array_copy_ctr, O3, O4); |
duke@435 | 2680 | __ retl(); |
duke@435 | 2681 | __ delayed()->mov(G0, O0); // return 0 |
duke@435 | 2682 | return start; |
duke@435 | 2683 | } |
duke@435 | 2684 | |
duke@435 | 2685 | |
duke@435 | 2686 | // Helper for generating a dynamic type check. |
duke@435 | 2687 | // Smashes only the given temp registers. |
duke@435 | 2688 | void generate_type_check(Register sub_klass, |
duke@435 | 2689 | Register super_check_offset, |
duke@435 | 2690 | Register super_klass, |
duke@435 | 2691 | Register temp, |
jrose@1079 | 2692 | Label& L_success) { |
duke@435 | 2693 | assert_different_registers(sub_klass, super_check_offset, super_klass, temp); |
duke@435 | 2694 | |
duke@435 | 2695 | BLOCK_COMMENT("type_check:"); |
duke@435 | 2696 | |
jrose@1079 | 2697 | Label L_miss, L_pop_to_miss; |
duke@435 | 2698 | |
duke@435 | 2699 | assert_clean_int(super_check_offset, temp); |
duke@435 | 2700 | |
jrose@1079 | 2701 | __ check_klass_subtype_fast_path(sub_klass, super_klass, temp, noreg, |
jrose@1079 | 2702 | &L_success, &L_miss, NULL, |
jrose@1079 | 2703 | super_check_offset); |
jrose@1079 | 2704 | |
jrose@1079 | 2705 | BLOCK_COMMENT("type_check_slow_path:"); |
duke@435 | 2706 | __ save_frame(0); |
jrose@1079 | 2707 | __ check_klass_subtype_slow_path(sub_klass->after_save(), |
jrose@1079 | 2708 | super_klass->after_save(), |
jrose@1079 | 2709 | L0, L1, L2, L4, |
jrose@1079 | 2710 | NULL, &L_pop_to_miss); |
kvn@3037 | 2711 | __ ba(L_success); |
jrose@1079 | 2712 | __ delayed()->restore(); |
jrose@1079 | 2713 | |
jrose@1079 | 2714 | __ bind(L_pop_to_miss); |
duke@435 | 2715 | __ restore(); |
duke@435 | 2716 | |
duke@435 | 2717 | // Fall through on failure! |
duke@435 | 2718 | __ BIND(L_miss); |
duke@435 | 2719 | } |
duke@435 | 2720 | |
duke@435 | 2721 | |
duke@435 | 2722 | // Generate stub for checked oop copy. |
duke@435 | 2723 | // |
duke@435 | 2724 | // Arguments for generated stub: |
duke@435 | 2725 | // from: O0 |
duke@435 | 2726 | // to: O1 |
duke@435 | 2727 | // count: O2 treated as signed |
duke@435 | 2728 | // ckoff: O3 (super_check_offset) |
duke@435 | 2729 | // ckval: O4 (super_klass) |
duke@435 | 2730 | // ret: O0 zero for success; (-1^K) where K is partial transfer count |
duke@435 | 2731 | // |
iveresov@2606 | 2732 | address generate_checkcast_copy(const char *name, address *entry, bool dest_uninitialized = false) { |
duke@435 | 2733 | |
duke@435 | 2734 | const Register O0_from = O0; // source array address |
duke@435 | 2735 | const Register O1_to = O1; // destination array address |
duke@435 | 2736 | const Register O2_count = O2; // elements count |
duke@435 | 2737 | const Register O3_ckoff = O3; // super_check_offset |
duke@435 | 2738 | const Register O4_ckval = O4; // super_klass |
duke@435 | 2739 | |
duke@435 | 2740 | const Register O5_offset = O5; // loop var, with stride wordSize |
duke@435 | 2741 | const Register G1_remain = G1; // loop var, with stride -1 |
duke@435 | 2742 | const Register G3_oop = G3; // actual oop copied |
duke@435 | 2743 | const Register G4_klass = G4; // oop._klass |
duke@435 | 2744 | const Register G5_super = G5; // oop._klass._primary_supers[ckval] |
duke@435 | 2745 | |
duke@435 | 2746 | __ align(CodeEntryAlignment); |
duke@435 | 2747 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2748 | address start = __ pc(); |
duke@435 | 2749 | |
duke@435 | 2750 | #ifdef ASSERT |
jrose@1079 | 2751 | // We sometimes save a frame (see generate_type_check below). |
duke@435 | 2752 | // If this will cause trouble, let's fail now instead of later. |
duke@435 | 2753 | __ save_frame(0); |
duke@435 | 2754 | __ restore(); |
duke@435 | 2755 | #endif |
duke@435 | 2756 | |
never@2199 | 2757 | assert_clean_int(O2_count, G1); // Make sure 'count' is clean int. |
never@2199 | 2758 | |
duke@435 | 2759 | #ifdef ASSERT |
duke@435 | 2760 | // caller guarantees that the arrays really are different |
duke@435 | 2761 | // otherwise, we would have to make conjoint checks |
duke@435 | 2762 | { Label L; |
duke@435 | 2763 | __ mov(O3, G1); // spill: overlap test smashes O3 |
duke@435 | 2764 | __ mov(O4, G4); // spill: overlap test smashes O4 |
coleenp@548 | 2765 | array_overlap_test(L, LogBytesPerHeapOop); |
duke@435 | 2766 | __ stop("checkcast_copy within a single array"); |
duke@435 | 2767 | __ bind(L); |
duke@435 | 2768 | __ mov(G1, O3); |
duke@435 | 2769 | __ mov(G4, O4); |
duke@435 | 2770 | } |
duke@435 | 2771 | #endif //ASSERT |
duke@435 | 2772 | |
iveresov@2595 | 2773 | if (entry != NULL) { |
iveresov@2595 | 2774 | *entry = __ pc(); |
iveresov@2595 | 2775 | // caller can pass a 64-bit byte count here (from generic stub) |
iveresov@2595 | 2776 | BLOCK_COMMENT("Entry:"); |
iveresov@2595 | 2777 | } |
iveresov@2606 | 2778 | gen_write_ref_array_pre_barrier(O1_to, O2_count, dest_uninitialized); |
duke@435 | 2779 | |
duke@435 | 2780 | Label load_element, store_element, do_card_marks, fail, done; |
duke@435 | 2781 | __ addcc(O2_count, 0, G1_remain); // initialize loop index, and test it |
duke@435 | 2782 | __ brx(Assembler::notZero, false, Assembler::pt, load_element); |
duke@435 | 2783 | __ delayed()->mov(G0, O5_offset); // offset from start of arrays |
duke@435 | 2784 | |
duke@435 | 2785 | // Empty array: Nothing to do. |
duke@435 | 2786 | inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr, O3, O4); |
duke@435 | 2787 | __ retl(); |
duke@435 | 2788 | __ delayed()->set(0, O0); // return 0 on (trivial) success |
duke@435 | 2789 | |
duke@435 | 2790 | // ======== begin loop ======== |
duke@435 | 2791 | // (Loop is rotated; its entry is load_element.) |
duke@435 | 2792 | // Loop variables: |
duke@435 | 2793 | // (O5 = 0; ; O5 += wordSize) --- offset from src, dest arrays |
duke@435 | 2794 | // (O2 = len; O2 != 0; O2--) --- number of oops *remaining* |
duke@435 | 2795 | // G3, G4, G5 --- current oop, oop.klass, oop.klass.super |
kvn@1800 | 2796 | __ align(OptoLoopAlignment); |
duke@435 | 2797 | |
jrose@1079 | 2798 | __ BIND(store_element); |
jrose@1079 | 2799 | __ deccc(G1_remain); // decrement the count |
coleenp@548 | 2800 | __ store_heap_oop(G3_oop, O1_to, O5_offset); // store the oop |
coleenp@548 | 2801 | __ inc(O5_offset, heapOopSize); // step to next offset |
duke@435 | 2802 | __ brx(Assembler::zero, true, Assembler::pt, do_card_marks); |
duke@435 | 2803 | __ delayed()->set(0, O0); // return -1 on success |
duke@435 | 2804 | |
duke@435 | 2805 | // ======== loop entry is here ======== |
jrose@1079 | 2806 | __ BIND(load_element); |
coleenp@548 | 2807 | __ load_heap_oop(O0_from, O5_offset, G3_oop); // load the oop |
kvn@3037 | 2808 | __ br_null_short(G3_oop, Assembler::pt, store_element); |
duke@435 | 2809 | |
coleenp@548 | 2810 | __ load_klass(G3_oop, G4_klass); // query the object klass |
duke@435 | 2811 | |
duke@435 | 2812 | generate_type_check(G4_klass, O3_ckoff, O4_ckval, G5_super, |
duke@435 | 2813 | // branch to this on success: |
jrose@1079 | 2814 | store_element); |
duke@435 | 2815 | // ======== end loop ======== |
duke@435 | 2816 | |
duke@435 | 2817 | // It was a real error; we must depend on the caller to finish the job. |
duke@435 | 2818 | // Register G1 has number of *remaining* oops, O2 number of *total* oops. |
duke@435 | 2819 | // Emit GC store barriers for the oops we have copied (O2 minus G1), |
duke@435 | 2820 | // and report their number to the caller. |
jrose@1079 | 2821 | __ BIND(fail); |
duke@435 | 2822 | __ subcc(O2_count, G1_remain, O2_count); |
duke@435 | 2823 | __ brx(Assembler::zero, false, Assembler::pt, done); |
duke@435 | 2824 | __ delayed()->not1(O2_count, O0); // report (-1^K) to caller |
duke@435 | 2825 | |
jrose@1079 | 2826 | __ BIND(do_card_marks); |
duke@435 | 2827 | gen_write_ref_array_post_barrier(O1_to, O2_count, O3); // store check on O1[0..O2] |
duke@435 | 2828 | |
jrose@1079 | 2829 | __ BIND(done); |
duke@435 | 2830 | inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr, O3, O4); |
duke@435 | 2831 | __ retl(); |
duke@435 | 2832 | __ delayed()->nop(); // return value in 00 |
duke@435 | 2833 | |
duke@435 | 2834 | return start; |
duke@435 | 2835 | } |
duke@435 | 2836 | |
duke@435 | 2837 | |
duke@435 | 2838 | // Generate 'unsafe' array copy stub |
duke@435 | 2839 | // Though just as safe as the other stubs, it takes an unscaled |
duke@435 | 2840 | // size_t argument instead of an element count. |
duke@435 | 2841 | // |
duke@435 | 2842 | // Arguments for generated stub: |
duke@435 | 2843 | // from: O0 |
duke@435 | 2844 | // to: O1 |
duke@435 | 2845 | // count: O2 byte count, treated as ssize_t, can be zero |
duke@435 | 2846 | // |
duke@435 | 2847 | // Examines the alignment of the operands and dispatches |
duke@435 | 2848 | // to a long, int, short, or byte copy loop. |
duke@435 | 2849 | // |
iveresov@2595 | 2850 | address generate_unsafe_copy(const char* name, |
iveresov@2595 | 2851 | address byte_copy_entry, |
iveresov@2595 | 2852 | address short_copy_entry, |
iveresov@2595 | 2853 | address int_copy_entry, |
iveresov@2595 | 2854 | address long_copy_entry) { |
duke@435 | 2855 | |
duke@435 | 2856 | const Register O0_from = O0; // source array address |
duke@435 | 2857 | const Register O1_to = O1; // destination array address |
duke@435 | 2858 | const Register O2_count = O2; // elements count |
duke@435 | 2859 | |
duke@435 | 2860 | const Register G1_bits = G1; // test copy of low bits |
duke@435 | 2861 | |
duke@435 | 2862 | __ align(CodeEntryAlignment); |
duke@435 | 2863 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2864 | address start = __ pc(); |
duke@435 | 2865 | |
duke@435 | 2866 | // bump this on entry, not on exit: |
duke@435 | 2867 | inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr, G1, G3); |
duke@435 | 2868 | |
duke@435 | 2869 | __ or3(O0_from, O1_to, G1_bits); |
duke@435 | 2870 | __ or3(O2_count, G1_bits, G1_bits); |
duke@435 | 2871 | |
duke@435 | 2872 | __ btst(BytesPerLong-1, G1_bits); |
duke@435 | 2873 | __ br(Assembler::zero, true, Assembler::pt, |
duke@435 | 2874 | long_copy_entry, relocInfo::runtime_call_type); |
duke@435 | 2875 | // scale the count on the way out: |
duke@435 | 2876 | __ delayed()->srax(O2_count, LogBytesPerLong, O2_count); |
duke@435 | 2877 | |
duke@435 | 2878 | __ btst(BytesPerInt-1, G1_bits); |
duke@435 | 2879 | __ br(Assembler::zero, true, Assembler::pt, |
duke@435 | 2880 | int_copy_entry, relocInfo::runtime_call_type); |
duke@435 | 2881 | // scale the count on the way out: |
duke@435 | 2882 | __ delayed()->srax(O2_count, LogBytesPerInt, O2_count); |
duke@435 | 2883 | |
duke@435 | 2884 | __ btst(BytesPerShort-1, G1_bits); |
duke@435 | 2885 | __ br(Assembler::zero, true, Assembler::pt, |
duke@435 | 2886 | short_copy_entry, relocInfo::runtime_call_type); |
duke@435 | 2887 | // scale the count on the way out: |
duke@435 | 2888 | __ delayed()->srax(O2_count, LogBytesPerShort, O2_count); |
duke@435 | 2889 | |
duke@435 | 2890 | __ br(Assembler::always, false, Assembler::pt, |
duke@435 | 2891 | byte_copy_entry, relocInfo::runtime_call_type); |
duke@435 | 2892 | __ delayed()->nop(); |
duke@435 | 2893 | |
duke@435 | 2894 | return start; |
duke@435 | 2895 | } |
duke@435 | 2896 | |
duke@435 | 2897 | |
duke@435 | 2898 | // Perform range checks on the proposed arraycopy. |
duke@435 | 2899 | // Kills the two temps, but nothing else. |
duke@435 | 2900 | // Also, clean the sign bits of src_pos and dst_pos. |
duke@435 | 2901 | void arraycopy_range_checks(Register src, // source array oop (O0) |
duke@435 | 2902 | Register src_pos, // source position (O1) |
duke@435 | 2903 | Register dst, // destination array oo (O2) |
duke@435 | 2904 | Register dst_pos, // destination position (O3) |
duke@435 | 2905 | Register length, // length of copy (O4) |
duke@435 | 2906 | Register temp1, Register temp2, |
duke@435 | 2907 | Label& L_failed) { |
duke@435 | 2908 | BLOCK_COMMENT("arraycopy_range_checks:"); |
duke@435 | 2909 | |
duke@435 | 2910 | // if (src_pos + length > arrayOop(src)->length() ) FAIL; |
duke@435 | 2911 | |
duke@435 | 2912 | const Register array_length = temp1; // scratch |
duke@435 | 2913 | const Register end_pos = temp2; // scratch |
duke@435 | 2914 | |
duke@435 | 2915 | // Note: This next instruction may be in the delay slot of a branch: |
duke@435 | 2916 | __ add(length, src_pos, end_pos); // src_pos + length |
duke@435 | 2917 | __ lduw(src, arrayOopDesc::length_offset_in_bytes(), array_length); |
duke@435 | 2918 | __ cmp(end_pos, array_length); |
duke@435 | 2919 | __ br(Assembler::greater, false, Assembler::pn, L_failed); |
duke@435 | 2920 | |
duke@435 | 2921 | // if (dst_pos + length > arrayOop(dst)->length() ) FAIL; |
duke@435 | 2922 | __ delayed()->add(length, dst_pos, end_pos); // dst_pos + length |
duke@435 | 2923 | __ lduw(dst, arrayOopDesc::length_offset_in_bytes(), array_length); |
duke@435 | 2924 | __ cmp(end_pos, array_length); |
duke@435 | 2925 | __ br(Assembler::greater, false, Assembler::pn, L_failed); |
duke@435 | 2926 | |
duke@435 | 2927 | // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'. |
duke@435 | 2928 | // Move with sign extension can be used since they are positive. |
duke@435 | 2929 | __ delayed()->signx(src_pos, src_pos); |
duke@435 | 2930 | __ signx(dst_pos, dst_pos); |
duke@435 | 2931 | |
duke@435 | 2932 | BLOCK_COMMENT("arraycopy_range_checks done"); |
duke@435 | 2933 | } |
duke@435 | 2934 | |
duke@435 | 2935 | |
duke@435 | 2936 | // |
duke@435 | 2937 | // Generate generic array copy stubs |
duke@435 | 2938 | // |
duke@435 | 2939 | // Input: |
duke@435 | 2940 | // O0 - src oop |
duke@435 | 2941 | // O1 - src_pos |
duke@435 | 2942 | // O2 - dst oop |
duke@435 | 2943 | // O3 - dst_pos |
duke@435 | 2944 | // O4 - element count |
duke@435 | 2945 | // |
duke@435 | 2946 | // Output: |
duke@435 | 2947 | // O0 == 0 - success |
duke@435 | 2948 | // O0 == -1 - need to call System.arraycopy |
duke@435 | 2949 | // |
iveresov@2595 | 2950 | address generate_generic_copy(const char *name, |
iveresov@2595 | 2951 | address entry_jbyte_arraycopy, |
iveresov@2595 | 2952 | address entry_jshort_arraycopy, |
iveresov@2595 | 2953 | address entry_jint_arraycopy, |
iveresov@2595 | 2954 | address entry_oop_arraycopy, |
iveresov@2595 | 2955 | address entry_jlong_arraycopy, |
iveresov@2595 | 2956 | address entry_checkcast_arraycopy) { |
duke@435 | 2957 | Label L_failed, L_objArray; |
duke@435 | 2958 | |
duke@435 | 2959 | // Input registers |
duke@435 | 2960 | const Register src = O0; // source array oop |
duke@435 | 2961 | const Register src_pos = O1; // source position |
duke@435 | 2962 | const Register dst = O2; // destination array oop |
duke@435 | 2963 | const Register dst_pos = O3; // destination position |
duke@435 | 2964 | const Register length = O4; // elements count |
duke@435 | 2965 | |
duke@435 | 2966 | // registers used as temp |
duke@435 | 2967 | const Register G3_src_klass = G3; // source array klass |
duke@435 | 2968 | const Register G4_dst_klass = G4; // destination array klass |
duke@435 | 2969 | const Register G5_lh = G5; // layout handler |
duke@435 | 2970 | const Register O5_temp = O5; |
duke@435 | 2971 | |
duke@435 | 2972 | __ align(CodeEntryAlignment); |
duke@435 | 2973 | StubCodeMark mark(this, "StubRoutines", name); |
duke@435 | 2974 | address start = __ pc(); |
duke@435 | 2975 | |
duke@435 | 2976 | // bump this on entry, not on exit: |
duke@435 | 2977 | inc_counter_np(SharedRuntime::_generic_array_copy_ctr, G1, G3); |
duke@435 | 2978 | |
duke@435 | 2979 | // In principle, the int arguments could be dirty. |
duke@435 | 2980 | //assert_clean_int(src_pos, G1); |
duke@435 | 2981 | //assert_clean_int(dst_pos, G1); |
duke@435 | 2982 | //assert_clean_int(length, G1); |
duke@435 | 2983 | |
duke@435 | 2984 | //----------------------------------------------------------------------- |
duke@435 | 2985 | // Assembler stubs will be used for this call to arraycopy |
duke@435 | 2986 | // if the following conditions are met: |
duke@435 | 2987 | // |
duke@435 | 2988 | // (1) src and dst must not be null. |
duke@435 | 2989 | // (2) src_pos must not be negative. |
duke@435 | 2990 | // (3) dst_pos must not be negative. |
duke@435 | 2991 | // (4) length must not be negative. |
duke@435 | 2992 | // (5) src klass and dst klass should be the same and not NULL. |
duke@435 | 2993 | // (6) src and dst should be arrays. |
duke@435 | 2994 | // (7) src_pos + length must not exceed length of src. |
duke@435 | 2995 | // (8) dst_pos + length must not exceed length of dst. |
duke@435 | 2996 | BLOCK_COMMENT("arraycopy initial argument checks"); |
duke@435 | 2997 | |
duke@435 | 2998 | // if (src == NULL) return -1; |
duke@435 | 2999 | __ br_null(src, false, Assembler::pn, L_failed); |
duke@435 | 3000 | |
duke@435 | 3001 | // if (src_pos < 0) return -1; |
duke@435 | 3002 | __ delayed()->tst(src_pos); |
duke@435 | 3003 | __ br(Assembler::negative, false, Assembler::pn, L_failed); |
duke@435 | 3004 | __ delayed()->nop(); |
duke@435 | 3005 | |
duke@435 | 3006 | // if (dst == NULL) return -1; |
duke@435 | 3007 | __ br_null(dst, false, Assembler::pn, L_failed); |
duke@435 | 3008 | |
duke@435 | 3009 | // if (dst_pos < 0) return -1; |
duke@435 | 3010 | __ delayed()->tst(dst_pos); |
duke@435 | 3011 | __ br(Assembler::negative, false, Assembler::pn, L_failed); |
duke@435 | 3012 | |
duke@435 | 3013 | // if (length < 0) return -1; |
duke@435 | 3014 | __ delayed()->tst(length); |
duke@435 | 3015 | __ br(Assembler::negative, false, Assembler::pn, L_failed); |
duke@435 | 3016 | |
duke@435 | 3017 | BLOCK_COMMENT("arraycopy argument klass checks"); |
duke@435 | 3018 | // get src->klass() |
coleenp@548 | 3019 | if (UseCompressedOops) { |
coleenp@548 | 3020 | __ delayed()->nop(); // ??? not good |
coleenp@548 | 3021 | __ load_klass(src, G3_src_klass); |
coleenp@548 | 3022 | } else { |
coleenp@548 | 3023 | __ delayed()->ld_ptr(src, oopDesc::klass_offset_in_bytes(), G3_src_klass); |
coleenp@548 | 3024 | } |
duke@435 | 3025 | |
duke@435 | 3026 | #ifdef ASSERT |
duke@435 | 3027 | // assert(src->klass() != NULL); |
duke@435 | 3028 | BLOCK_COMMENT("assert klasses not null"); |
duke@435 | 3029 | { Label L_a, L_b; |
kvn@3037 | 3030 | __ br_notnull_short(G3_src_klass, Assembler::pt, L_b); // it is broken if klass is NULL |
duke@435 | 3031 | __ bind(L_a); |
duke@435 | 3032 | __ stop("broken null klass"); |
duke@435 | 3033 | __ bind(L_b); |
coleenp@548 | 3034 | __ load_klass(dst, G4_dst_klass); |
duke@435 | 3035 | __ br_null(G4_dst_klass, false, Assembler::pn, L_a); // this would be broken also |
duke@435 | 3036 | __ delayed()->mov(G0, G4_dst_klass); // scribble the temp |
duke@435 | 3037 | BLOCK_COMMENT("assert done"); |
duke@435 | 3038 | } |
duke@435 | 3039 | #endif |
duke@435 | 3040 | |
duke@435 | 3041 | // Load layout helper |
duke@435 | 3042 | // |
duke@435 | 3043 | // |array_tag| | header_size | element_type | |log2_element_size| |
duke@435 | 3044 | // 32 30 24 16 8 2 0 |
duke@435 | 3045 | // |
duke@435 | 3046 | // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 |
duke@435 | 3047 | // |
duke@435 | 3048 | |
duke@435 | 3049 | int lh_offset = klassOopDesc::header_size() * HeapWordSize + |
duke@435 | 3050 | Klass::layout_helper_offset_in_bytes(); |
duke@435 | 3051 | |
duke@435 | 3052 | // Load 32-bits signed value. Use br() instruction with it to check icc. |
duke@435 | 3053 | __ lduw(G3_src_klass, lh_offset, G5_lh); |
duke@435 | 3054 | |
coleenp@548 | 3055 | if (UseCompressedOops) { |
coleenp@548 | 3056 | __ load_klass(dst, G4_dst_klass); |
coleenp@548 | 3057 | } |
duke@435 | 3058 | // Handle objArrays completely differently... |
duke@435 | 3059 | juint objArray_lh = Klass::array_layout_helper(T_OBJECT); |
duke@435 | 3060 | __ set(objArray_lh, O5_temp); |
duke@435 | 3061 | __ cmp(G5_lh, O5_temp); |
duke@435 | 3062 | __ br(Assembler::equal, false, Assembler::pt, L_objArray); |
coleenp@548 | 3063 | if (UseCompressedOops) { |
coleenp@548 | 3064 | __ delayed()->nop(); |
coleenp@548 | 3065 | } else { |
coleenp@548 | 3066 | __ delayed()->ld_ptr(dst, oopDesc::klass_offset_in_bytes(), G4_dst_klass); |
coleenp@548 | 3067 | } |
duke@435 | 3068 | |
duke@435 | 3069 | // if (src->klass() != dst->klass()) return -1; |
kvn@3037 | 3070 | __ cmp_and_brx_short(G3_src_klass, G4_dst_klass, Assembler::notEqual, Assembler::pn, L_failed); |
duke@435 | 3071 | |
duke@435 | 3072 | // if (!src->is_Array()) return -1; |
duke@435 | 3073 | __ cmp(G5_lh, Klass::_lh_neutral_value); // < 0 |
duke@435 | 3074 | __ br(Assembler::greaterEqual, false, Assembler::pn, L_failed); |
duke@435 | 3075 | |
duke@435 | 3076 | // At this point, it is known to be a typeArray (array_tag 0x3). |
duke@435 | 3077 | #ifdef ASSERT |
duke@435 | 3078 | __ delayed()->nop(); |
duke@435 | 3079 | { Label L; |
duke@435 | 3080 | jint lh_prim_tag_in_place = (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift); |
duke@435 | 3081 | __ set(lh_prim_tag_in_place, O5_temp); |
duke@435 | 3082 | __ cmp(G5_lh, O5_temp); |
duke@435 | 3083 | __ br(Assembler::greaterEqual, false, Assembler::pt, L); |
duke@435 | 3084 | __ delayed()->nop(); |
duke@435 | 3085 | __ stop("must be a primitive array"); |
duke@435 | 3086 | __ bind(L); |
duke@435 | 3087 | } |
duke@435 | 3088 | #else |
duke@435 | 3089 | __ delayed(); // match next insn to prev branch |
duke@435 | 3090 | #endif |
duke@435 | 3091 | |
duke@435 | 3092 | arraycopy_range_checks(src, src_pos, dst, dst_pos, length, |
duke@435 | 3093 | O5_temp, G4_dst_klass, L_failed); |
duke@435 | 3094 | |
duke@435 | 3095 | // typeArrayKlass |
duke@435 | 3096 | // |
duke@435 | 3097 | // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); |
duke@435 | 3098 | // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); |
duke@435 | 3099 | // |
duke@435 | 3100 | |
duke@435 | 3101 | const Register G4_offset = G4_dst_klass; // array offset |
duke@435 | 3102 | const Register G3_elsize = G3_src_klass; // log2 element size |
duke@435 | 3103 | |
duke@435 | 3104 | __ srl(G5_lh, Klass::_lh_header_size_shift, G4_offset); |
duke@435 | 3105 | __ and3(G4_offset, Klass::_lh_header_size_mask, G4_offset); // array_offset |
duke@435 | 3106 | __ add(src, G4_offset, src); // src array offset |
duke@435 | 3107 | __ add(dst, G4_offset, dst); // dst array offset |
duke@435 | 3108 | __ and3(G5_lh, Klass::_lh_log2_element_size_mask, G3_elsize); // log2 element size |
duke@435 | 3109 | |
duke@435 | 3110 | // next registers should be set before the jump to corresponding stub |
duke@435 | 3111 | const Register from = O0; // source array address |
duke@435 | 3112 | const Register to = O1; // destination array address |
duke@435 | 3113 | const Register count = O2; // elements count |
duke@435 | 3114 | |
duke@435 | 3115 | // 'from', 'to', 'count' registers should be set in this order |
duke@435 | 3116 | // since they are the same as 'src', 'src_pos', 'dst'. |
duke@435 | 3117 | |
duke@435 | 3118 | BLOCK_COMMENT("scale indexes to element size"); |
duke@435 | 3119 | __ sll_ptr(src_pos, G3_elsize, src_pos); |
duke@435 | 3120 | __ sll_ptr(dst_pos, G3_elsize, dst_pos); |
duke@435 | 3121 | __ add(src, src_pos, from); // src_addr |
duke@435 | 3122 | __ add(dst, dst_pos, to); // dst_addr |
duke@435 | 3123 | |
duke@435 | 3124 | BLOCK_COMMENT("choose copy loop based on element size"); |
duke@435 | 3125 | __ cmp(G3_elsize, 0); |
iveresov@2595 | 3126 | __ br(Assembler::equal, true, Assembler::pt, entry_jbyte_arraycopy); |
duke@435 | 3127 | __ delayed()->signx(length, count); // length |
duke@435 | 3128 | |
duke@435 | 3129 | __ cmp(G3_elsize, LogBytesPerShort); |
iveresov@2595 | 3130 | __ br(Assembler::equal, true, Assembler::pt, entry_jshort_arraycopy); |
duke@435 | 3131 | __ delayed()->signx(length, count); // length |
duke@435 | 3132 | |
duke@435 | 3133 | __ cmp(G3_elsize, LogBytesPerInt); |
iveresov@2595 | 3134 | __ br(Assembler::equal, true, Assembler::pt, entry_jint_arraycopy); |
duke@435 | 3135 | __ delayed()->signx(length, count); // length |
duke@435 | 3136 | #ifdef ASSERT |
duke@435 | 3137 | { Label L; |
kvn@3037 | 3138 | __ cmp_and_br_short(G3_elsize, LogBytesPerLong, Assembler::equal, Assembler::pt, L); |
duke@435 | 3139 | __ stop("must be long copy, but elsize is wrong"); |
duke@435 | 3140 | __ bind(L); |
duke@435 | 3141 | } |
duke@435 | 3142 | #endif |
iveresov@2595 | 3143 | __ br(Assembler::always, false, Assembler::pt, entry_jlong_arraycopy); |
duke@435 | 3144 | __ delayed()->signx(length, count); // length |
duke@435 | 3145 | |
duke@435 | 3146 | // objArrayKlass |
duke@435 | 3147 | __ BIND(L_objArray); |
duke@435 | 3148 | // live at this point: G3_src_klass, G4_dst_klass, src[_pos], dst[_pos], length |
duke@435 | 3149 | |
duke@435 | 3150 | Label L_plain_copy, L_checkcast_copy; |
duke@435 | 3151 | // test array classes for subtyping |
duke@435 | 3152 | __ cmp(G3_src_klass, G4_dst_klass); // usual case is exact equality |
duke@435 | 3153 | __ brx(Assembler::notEqual, true, Assembler::pn, L_checkcast_copy); |
duke@435 | 3154 | __ delayed()->lduw(G4_dst_klass, lh_offset, O5_temp); // hoisted from below |
duke@435 | 3155 | |
duke@435 | 3156 | // Identically typed arrays can be copied without element-wise checks. |
duke@435 | 3157 | arraycopy_range_checks(src, src_pos, dst, dst_pos, length, |
duke@435 | 3158 | O5_temp, G5_lh, L_failed); |
duke@435 | 3159 | |
duke@435 | 3160 | __ add(src, arrayOopDesc::base_offset_in_bytes(T_OBJECT), src); //src offset |
duke@435 | 3161 | __ add(dst, arrayOopDesc::base_offset_in_bytes(T_OBJECT), dst); //dst offset |
coleenp@548 | 3162 | __ sll_ptr(src_pos, LogBytesPerHeapOop, src_pos); |
coleenp@548 | 3163 | __ sll_ptr(dst_pos, LogBytesPerHeapOop, dst_pos); |
duke@435 | 3164 | __ add(src, src_pos, from); // src_addr |
duke@435 | 3165 | __ add(dst, dst_pos, to); // dst_addr |
duke@435 | 3166 | __ BIND(L_plain_copy); |
iveresov@2595 | 3167 | __ br(Assembler::always, false, Assembler::pt, entry_oop_arraycopy); |
duke@435 | 3168 | __ delayed()->signx(length, count); // length |
duke@435 | 3169 | |
duke@435 | 3170 | __ BIND(L_checkcast_copy); |
duke@435 | 3171 | // live at this point: G3_src_klass, G4_dst_klass |
duke@435 | 3172 | { |
duke@435 | 3173 | // Before looking at dst.length, make sure dst is also an objArray. |
duke@435 | 3174 | // lduw(G4_dst_klass, lh_offset, O5_temp); // hoisted to delay slot |
duke@435 | 3175 | __ cmp(G5_lh, O5_temp); |
duke@435 | 3176 | __ br(Assembler::notEqual, false, Assembler::pn, L_failed); |
duke@435 | 3177 | |
duke@435 | 3178 | // It is safe to examine both src.length and dst.length. |
duke@435 | 3179 | __ delayed(); // match next insn to prev branch |
duke@435 | 3180 | arraycopy_range_checks(src, src_pos, dst, dst_pos, length, |
duke@435 | 3181 | O5_temp, G5_lh, L_failed); |
duke@435 | 3182 | |
duke@435 | 3183 | // Marshal the base address arguments now, freeing registers. |
duke@435 | 3184 | __ add(src, arrayOopDesc::base_offset_in_bytes(T_OBJECT), src); //src offset |
duke@435 | 3185 | __ add(dst, arrayOopDesc::base_offset_in_bytes(T_OBJECT), dst); //dst offset |
coleenp@548 | 3186 | __ sll_ptr(src_pos, LogBytesPerHeapOop, src_pos); |
coleenp@548 | 3187 | __ sll_ptr(dst_pos, LogBytesPerHeapOop, dst_pos); |
duke@435 | 3188 | __ add(src, src_pos, from); // src_addr |
duke@435 | 3189 | __ add(dst, dst_pos, to); // dst_addr |
duke@435 | 3190 | __ signx(length, count); // length (reloaded) |
duke@435 | 3191 | |
duke@435 | 3192 | Register sco_temp = O3; // this register is free now |
duke@435 | 3193 | assert_different_registers(from, to, count, sco_temp, |
duke@435 | 3194 | G4_dst_klass, G3_src_klass); |
duke@435 | 3195 | |
duke@435 | 3196 | // Generate the type check. |
duke@435 | 3197 | int sco_offset = (klassOopDesc::header_size() * HeapWordSize + |
duke@435 | 3198 | Klass::super_check_offset_offset_in_bytes()); |
duke@435 | 3199 | __ lduw(G4_dst_klass, sco_offset, sco_temp); |
duke@435 | 3200 | generate_type_check(G3_src_klass, sco_temp, G4_dst_klass, |
duke@435 | 3201 | O5_temp, L_plain_copy); |
duke@435 | 3202 | |
duke@435 | 3203 | // Fetch destination element klass from the objArrayKlass header. |
duke@435 | 3204 | int ek_offset = (klassOopDesc::header_size() * HeapWordSize + |
duke@435 | 3205 | objArrayKlass::element_klass_offset_in_bytes()); |
duke@435 | 3206 | |
duke@435 | 3207 | // the checkcast_copy loop needs two extra arguments: |
duke@435 | 3208 | __ ld_ptr(G4_dst_klass, ek_offset, O4); // dest elem klass |
duke@435 | 3209 | // lduw(O4, sco_offset, O3); // sco of elem klass |
duke@435 | 3210 | |
iveresov@2595 | 3211 | __ br(Assembler::always, false, Assembler::pt, entry_checkcast_arraycopy); |
duke@435 | 3212 | __ delayed()->lduw(O4, sco_offset, O3); |
duke@435 | 3213 | } |
duke@435 | 3214 | |
duke@435 | 3215 | __ BIND(L_failed); |
duke@435 | 3216 | __ retl(); |
duke@435 | 3217 | __ delayed()->sub(G0, 1, O0); // return -1 |
duke@435 | 3218 | return start; |
duke@435 | 3219 | } |
duke@435 | 3220 | |
kvn@3092 | 3221 | // |
kvn@3092 | 3222 | // Generate stub for heap zeroing. |
kvn@3092 | 3223 | // "to" address is aligned to jlong (8 bytes). |
kvn@3092 | 3224 | // |
kvn@3092 | 3225 | // Arguments for generated stub: |
kvn@3092 | 3226 | // to: O0 |
kvn@3092 | 3227 | // count: O1 treated as signed (count of HeapWord) |
kvn@3092 | 3228 | // count could be 0 |
kvn@3092 | 3229 | // |
kvn@3092 | 3230 | address generate_zero_aligned_words(const char* name) { |
kvn@3092 | 3231 | __ align(CodeEntryAlignment); |
kvn@3092 | 3232 | StubCodeMark mark(this, "StubRoutines", name); |
kvn@3092 | 3233 | address start = __ pc(); |
kvn@3092 | 3234 | |
kvn@3092 | 3235 | const Register to = O0; // source array address |
kvn@3092 | 3236 | const Register count = O1; // HeapWords count |
kvn@3092 | 3237 | const Register temp = O2; // scratch |
kvn@3092 | 3238 | |
kvn@3092 | 3239 | Label Ldone; |
kvn@3092 | 3240 | __ sllx(count, LogHeapWordSize, count); // to bytes count |
kvn@3092 | 3241 | // Use BIS for zeroing |
kvn@3092 | 3242 | __ bis_zeroing(to, count, temp, Ldone); |
kvn@3092 | 3243 | __ bind(Ldone); |
kvn@3092 | 3244 | __ retl(); |
kvn@3092 | 3245 | __ delayed()->nop(); |
kvn@3092 | 3246 | return start; |
kvn@3092 | 3247 | } |
kvn@3092 | 3248 | |
duke@435 | 3249 | void generate_arraycopy_stubs() { |
iveresov@2595 | 3250 | address entry; |
iveresov@2595 | 3251 | address entry_jbyte_arraycopy; |
iveresov@2595 | 3252 | address entry_jshort_arraycopy; |
iveresov@2595 | 3253 | address entry_jint_arraycopy; |
iveresov@2595 | 3254 | address entry_oop_arraycopy; |
iveresov@2595 | 3255 | address entry_jlong_arraycopy; |
iveresov@2595 | 3256 | address entry_checkcast_arraycopy; |
iveresov@2595 | 3257 | |
iveresov@2606 | 3258 | //*** jbyte |
iveresov@2606 | 3259 | // Always need aligned and unaligned versions |
iveresov@2606 | 3260 | StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry, |
iveresov@2606 | 3261 | "jbyte_disjoint_arraycopy"); |
iveresov@2606 | 3262 | StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, |
iveresov@2606 | 3263 | &entry_jbyte_arraycopy, |
iveresov@2606 | 3264 | "jbyte_arraycopy"); |
iveresov@2606 | 3265 | StubRoutines::_arrayof_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(true, &entry, |
iveresov@2606 | 3266 | "arrayof_jbyte_disjoint_arraycopy"); |
iveresov@2606 | 3267 | StubRoutines::_arrayof_jbyte_arraycopy = generate_conjoint_byte_copy(true, entry, NULL, |
iveresov@2606 | 3268 | "arrayof_jbyte_arraycopy"); |
iveresov@2606 | 3269 | |
iveresov@2606 | 3270 | //*** jshort |
iveresov@2606 | 3271 | // Always need aligned and unaligned versions |
iveresov@2606 | 3272 | StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry, |
iveresov@2606 | 3273 | "jshort_disjoint_arraycopy"); |
iveresov@2606 | 3274 | StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, |
iveresov@2606 | 3275 | &entry_jshort_arraycopy, |
iveresov@2606 | 3276 | "jshort_arraycopy"); |
iveresov@2595 | 3277 | StubRoutines::_arrayof_jshort_disjoint_arraycopy = generate_disjoint_short_copy(true, &entry, |
iveresov@2595 | 3278 | "arrayof_jshort_disjoint_arraycopy"); |
iveresov@2595 | 3279 | StubRoutines::_arrayof_jshort_arraycopy = generate_conjoint_short_copy(true, entry, NULL, |
iveresov@2595 | 3280 | "arrayof_jshort_arraycopy"); |
iveresov@2595 | 3281 | |
iveresov@2606 | 3282 | //*** jint |
iveresov@2606 | 3283 | // Aligned versions |
iveresov@2606 | 3284 | StubRoutines::_arrayof_jint_disjoint_arraycopy = generate_disjoint_int_copy(true, &entry, |
iveresov@2606 | 3285 | "arrayof_jint_disjoint_arraycopy"); |
iveresov@2606 | 3286 | StubRoutines::_arrayof_jint_arraycopy = generate_conjoint_int_copy(true, entry, &entry_jint_arraycopy, |
iveresov@2606 | 3287 | "arrayof_jint_arraycopy"); |
duke@435 | 3288 | #ifdef _LP64 |
iveresov@2606 | 3289 | // In 64 bit we need both aligned and unaligned versions of jint arraycopy. |
iveresov@2606 | 3290 | // entry_jint_arraycopy always points to the unaligned version (notice that we overwrite it). |
iveresov@2606 | 3291 | StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_copy(false, &entry, |
iveresov@2606 | 3292 | "jint_disjoint_arraycopy"); |
iveresov@2606 | 3293 | StubRoutines::_jint_arraycopy = generate_conjoint_int_copy(false, entry, |
iveresov@2606 | 3294 | &entry_jint_arraycopy, |
iveresov@2606 | 3295 | "jint_arraycopy"); |
iveresov@2606 | 3296 | #else |
iveresov@2606 | 3297 | // In 32 bit jints are always HeapWordSize aligned, so always use the aligned version |
iveresov@2606 | 3298 | // (in fact in 32bit we always have a pre-loop part even in the aligned version, |
iveresov@2606 | 3299 | // because it uses 64-bit loads/stores, so the aligned flag is actually ignored). |
iveresov@2606 | 3300 | StubRoutines::_jint_disjoint_arraycopy = StubRoutines::_arrayof_jint_disjoint_arraycopy; |
iveresov@2606 | 3301 | StubRoutines::_jint_arraycopy = StubRoutines::_arrayof_jint_arraycopy; |
duke@435 | 3302 | #endif |
iveresov@2595 | 3303 | |
iveresov@2606 | 3304 | |
iveresov@2606 | 3305 | //*** jlong |
iveresov@2606 | 3306 | // It is always aligned |
iveresov@2606 | 3307 | StubRoutines::_arrayof_jlong_disjoint_arraycopy = generate_disjoint_long_copy(true, &entry, |
iveresov@2606 | 3308 | "arrayof_jlong_disjoint_arraycopy"); |
iveresov@2606 | 3309 | StubRoutines::_arrayof_jlong_arraycopy = generate_conjoint_long_copy(true, entry, &entry_jlong_arraycopy, |
iveresov@2606 | 3310 | "arrayof_jlong_arraycopy"); |
iveresov@2606 | 3311 | StubRoutines::_jlong_disjoint_arraycopy = StubRoutines::_arrayof_jlong_disjoint_arraycopy; |
iveresov@2606 | 3312 | StubRoutines::_jlong_arraycopy = StubRoutines::_arrayof_jlong_arraycopy; |
iveresov@2606 | 3313 | |
iveresov@2606 | 3314 | |
iveresov@2606 | 3315 | //*** oops |
iveresov@2606 | 3316 | // Aligned versions |
iveresov@2606 | 3317 | StubRoutines::_arrayof_oop_disjoint_arraycopy = generate_disjoint_oop_copy(true, &entry, |
iveresov@2606 | 3318 | "arrayof_oop_disjoint_arraycopy"); |
iveresov@2606 | 3319 | StubRoutines::_arrayof_oop_arraycopy = generate_conjoint_oop_copy(true, entry, &entry_oop_arraycopy, |
iveresov@2606 | 3320 | "arrayof_oop_arraycopy"); |
iveresov@2606 | 3321 | // Aligned versions without pre-barriers |
iveresov@2606 | 3322 | StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = generate_disjoint_oop_copy(true, &entry, |
iveresov@2606 | 3323 | "arrayof_oop_disjoint_arraycopy_uninit", |
iveresov@2606 | 3324 | /*dest_uninitialized*/true); |
iveresov@2606 | 3325 | StubRoutines::_arrayof_oop_arraycopy_uninit = generate_conjoint_oop_copy(true, entry, NULL, |
iveresov@2606 | 3326 | "arrayof_oop_arraycopy_uninit", |
iveresov@2606 | 3327 | /*dest_uninitialized*/true); |
iveresov@2606 | 3328 | #ifdef _LP64 |
iveresov@2606 | 3329 | if (UseCompressedOops) { |
iveresov@2606 | 3330 | // With compressed oops we need unaligned versions, notice that we overwrite entry_oop_arraycopy. |
iveresov@2606 | 3331 | StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_oop_copy(false, &entry, |
iveresov@2606 | 3332 | "oop_disjoint_arraycopy"); |
iveresov@2606 | 3333 | StubRoutines::_oop_arraycopy = generate_conjoint_oop_copy(false, entry, &entry_oop_arraycopy, |
iveresov@2606 | 3334 | "oop_arraycopy"); |
iveresov@2606 | 3335 | // Unaligned versions without pre-barriers |
iveresov@2606 | 3336 | StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_oop_copy(false, &entry, |
iveresov@2606 | 3337 | "oop_disjoint_arraycopy_uninit", |
iveresov@2606 | 3338 | /*dest_uninitialized*/true); |
iveresov@2606 | 3339 | StubRoutines::_oop_arraycopy_uninit = generate_conjoint_oop_copy(false, entry, NULL, |
iveresov@2606 | 3340 | "oop_arraycopy_uninit", |
iveresov@2606 | 3341 | /*dest_uninitialized*/true); |
iveresov@2606 | 3342 | } else |
iveresov@2606 | 3343 | #endif |
iveresov@2606 | 3344 | { |
iveresov@2606 | 3345 | // oop arraycopy is always aligned on 32bit and 64bit without compressed oops |
iveresov@2606 | 3346 | StubRoutines::_oop_disjoint_arraycopy = StubRoutines::_arrayof_oop_disjoint_arraycopy; |
iveresov@2606 | 3347 | StubRoutines::_oop_arraycopy = StubRoutines::_arrayof_oop_arraycopy; |
iveresov@2606 | 3348 | StubRoutines::_oop_disjoint_arraycopy_uninit = StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit; |
iveresov@2606 | 3349 | StubRoutines::_oop_arraycopy_uninit = StubRoutines::_arrayof_oop_arraycopy_uninit; |
iveresov@2606 | 3350 | } |
iveresov@2606 | 3351 | |
iveresov@2606 | 3352 | StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy); |
iveresov@2606 | 3353 | StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, |
iveresov@2606 | 3354 | /*dest_uninitialized*/true); |
iveresov@2606 | 3355 | |
iveresov@2595 | 3356 | StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy", |
iveresov@2595 | 3357 | entry_jbyte_arraycopy, |
iveresov@2595 | 3358 | entry_jshort_arraycopy, |
iveresov@2595 | 3359 | entry_jint_arraycopy, |
iveresov@2595 | 3360 | entry_jlong_arraycopy); |
iveresov@2595 | 3361 | StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy", |
iveresov@2595 | 3362 | entry_jbyte_arraycopy, |
iveresov@2595 | 3363 | entry_jshort_arraycopy, |
iveresov@2595 | 3364 | entry_jint_arraycopy, |
iveresov@2595 | 3365 | entry_oop_arraycopy, |
iveresov@2595 | 3366 | entry_jlong_arraycopy, |
iveresov@2595 | 3367 | entry_checkcast_arraycopy); |
never@2118 | 3368 | |
never@2118 | 3369 | StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill"); |
never@2118 | 3370 | StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill"); |
never@2118 | 3371 | StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill"); |
never@2118 | 3372 | StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill"); |
never@2118 | 3373 | StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill"); |
never@2118 | 3374 | StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill"); |
kvn@3092 | 3375 | |
kvn@3092 | 3376 | if (UseBlockZeroing) { |
kvn@3092 | 3377 | StubRoutines::_zero_aligned_words = generate_zero_aligned_words("zero_aligned_words"); |
kvn@3092 | 3378 | } |
duke@435 | 3379 | } |
duke@435 | 3380 | |
duke@435 | 3381 | void generate_initial() { |
duke@435 | 3382 | // Generates all stubs and initializes the entry points |
duke@435 | 3383 | |
duke@435 | 3384 | //------------------------------------------------------------------------------------------------------------------------ |
duke@435 | 3385 | // entry points that exist in all platforms |
duke@435 | 3386 | // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than |
duke@435 | 3387 | // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp. |
duke@435 | 3388 | StubRoutines::_forward_exception_entry = generate_forward_exception(); |
duke@435 | 3389 | |
duke@435 | 3390 | StubRoutines::_call_stub_entry = generate_call_stub(StubRoutines::_call_stub_return_address); |
duke@435 | 3391 | StubRoutines::_catch_exception_entry = generate_catch_exception(); |
duke@435 | 3392 | |
duke@435 | 3393 | //------------------------------------------------------------------------------------------------------------------------ |
duke@435 | 3394 | // entry points that are platform specific |
duke@435 | 3395 | StubRoutines::Sparc::_test_stop_entry = generate_test_stop(); |
duke@435 | 3396 | |
duke@435 | 3397 | StubRoutines::Sparc::_stop_subroutine_entry = generate_stop_subroutine(); |
duke@435 | 3398 | StubRoutines::Sparc::_flush_callers_register_windows_entry = generate_flush_callers_register_windows(); |
duke@435 | 3399 | |
duke@435 | 3400 | #if !defined(COMPILER2) && !defined(_LP64) |
duke@435 | 3401 | StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); |
duke@435 | 3402 | StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg(); |
duke@435 | 3403 | StubRoutines::_atomic_add_entry = generate_atomic_add(); |
duke@435 | 3404 | StubRoutines::_atomic_xchg_ptr_entry = StubRoutines::_atomic_xchg_entry; |
duke@435 | 3405 | StubRoutines::_atomic_cmpxchg_ptr_entry = StubRoutines::_atomic_cmpxchg_entry; |
duke@435 | 3406 | StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long(); |
duke@435 | 3407 | StubRoutines::_atomic_add_ptr_entry = StubRoutines::_atomic_add_entry; |
duke@435 | 3408 | #endif // COMPILER2 !=> _LP64 |
never@2978 | 3409 | |
never@2978 | 3410 | // Build this early so it's available for the interpreter. The |
never@2978 | 3411 | // stub expects the required and actual type to already be in O1 |
never@2978 | 3412 | // and O2 respectively. |
never@2978 | 3413 | StubRoutines::_throw_WrongMethodTypeException_entry = |
never@2978 | 3414 | generate_throw_exception("WrongMethodTypeException throw_exception", |
never@2978 | 3415 | CAST_FROM_FN_PTR(address, SharedRuntime::throw_WrongMethodTypeException), |
never@3136 | 3416 | G5_method_type, G3_method_handle); |
duke@435 | 3417 | } |
duke@435 | 3418 | |
duke@435 | 3419 | |
duke@435 | 3420 | void generate_all() { |
duke@435 | 3421 | // Generates all stubs and initializes the entry points |
duke@435 | 3422 | |
kvn@1077 | 3423 | // Generate partial_subtype_check first here since its code depends on |
kvn@1077 | 3424 | // UseZeroBaseCompressedOops which is defined after heap initialization. |
kvn@1077 | 3425 | StubRoutines::Sparc::_partial_subtype_check = generate_partial_subtype_check(); |
duke@435 | 3426 | // These entry points require SharedInfo::stack0 to be set up in non-core builds |
never@3136 | 3427 | StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError)); |
never@3136 | 3428 | StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError)); |
never@3136 | 3429 | StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call)); |
never@3136 | 3430 | StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError)); |
duke@435 | 3431 | |
duke@435 | 3432 | StubRoutines::_handler_for_unsafe_access_entry = |
duke@435 | 3433 | generate_handler_for_unsafe_access(); |
duke@435 | 3434 | |
duke@435 | 3435 | // support for verify_oop (must happen after universe_init) |
duke@435 | 3436 | StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop_subroutine(); |
duke@435 | 3437 | |
duke@435 | 3438 | // arraycopy stubs used by compilers |
duke@435 | 3439 | generate_arraycopy_stubs(); |
never@1609 | 3440 | |
never@1609 | 3441 | // Don't initialize the platform math functions since sparc |
never@1609 | 3442 | // doesn't have intrinsics for these operations. |
duke@435 | 3443 | } |
duke@435 | 3444 | |
duke@435 | 3445 | |
duke@435 | 3446 | public: |
duke@435 | 3447 | StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { |
duke@435 | 3448 | // replace the standard masm with a special one: |
duke@435 | 3449 | _masm = new MacroAssembler(code); |
duke@435 | 3450 | |
duke@435 | 3451 | _stub_count = !all ? 0x100 : 0x200; |
duke@435 | 3452 | if (all) { |
duke@435 | 3453 | generate_all(); |
duke@435 | 3454 | } else { |
duke@435 | 3455 | generate_initial(); |
duke@435 | 3456 | } |
duke@435 | 3457 | |
duke@435 | 3458 | // make sure this stub is available for all local calls |
duke@435 | 3459 | if (_atomic_add_stub.is_unbound()) { |
duke@435 | 3460 | // generate a second time, if necessary |
duke@435 | 3461 | (void) generate_atomic_add(); |
duke@435 | 3462 | } |
duke@435 | 3463 | } |
duke@435 | 3464 | |
duke@435 | 3465 | |
duke@435 | 3466 | private: |
duke@435 | 3467 | int _stub_count; |
duke@435 | 3468 | void stub_prolog(StubCodeDesc* cdesc) { |
duke@435 | 3469 | # ifdef ASSERT |
duke@435 | 3470 | // put extra information in the stub code, to make it more readable |
duke@435 | 3471 | #ifdef _LP64 |
duke@435 | 3472 | // Write the high part of the address |
duke@435 | 3473 | // [RGV] Check if there is a dependency on the size of this prolog |
duke@435 | 3474 | __ emit_data((intptr_t)cdesc >> 32, relocInfo::none); |
duke@435 | 3475 | #endif |
duke@435 | 3476 | __ emit_data((intptr_t)cdesc, relocInfo::none); |
duke@435 | 3477 | __ emit_data(++_stub_count, relocInfo::none); |
duke@435 | 3478 | # endif |
duke@435 | 3479 | align(true); |
duke@435 | 3480 | } |
duke@435 | 3481 | |
duke@435 | 3482 | void align(bool at_header = false) { |
duke@435 | 3483 | // %%%%% move this constant somewhere else |
duke@435 | 3484 | // UltraSPARC cache line size is 8 instructions: |
duke@435 | 3485 | const unsigned int icache_line_size = 32; |
duke@435 | 3486 | const unsigned int icache_half_line_size = 16; |
duke@435 | 3487 | |
duke@435 | 3488 | if (at_header) { |
duke@435 | 3489 | while ((intptr_t)(__ pc()) % icache_line_size != 0) { |
duke@435 | 3490 | __ emit_data(0, relocInfo::none); |
duke@435 | 3491 | } |
duke@435 | 3492 | } else { |
duke@435 | 3493 | while ((intptr_t)(__ pc()) % icache_half_line_size != 0) { |
duke@435 | 3494 | __ nop(); |
duke@435 | 3495 | } |
duke@435 | 3496 | } |
duke@435 | 3497 | } |
duke@435 | 3498 | |
duke@435 | 3499 | }; // end class declaration |
duke@435 | 3500 | |
duke@435 | 3501 | void StubGenerator_generate(CodeBuffer* code, bool all) { |
duke@435 | 3502 | StubGenerator g(code, all); |
duke@435 | 3503 | } |