Mercurial > hg > truffle
annotate src/cpu/sparc/vm/c1_MacroAssembler_sparc.cpp @ 2100:b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
Summary: Track allocated bytes in Thread's, update on TLAB retirement and direct allocation in Eden and tenured, add JNI methods for ThreadMXBean.
Reviewed-by: coleenp, kvn, dholmes, ysr
| author | phh |
|---|---|
| date | Fri, 07 Jan 2011 10:42:32 -0500 |
| parents | ac637b7220d1 |
| children | 5577848f5923 |
| rev | line source |
|---|---|
| 0 | 1 /* |
|
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1295
diff
changeset
|
2 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved. |
| 0 | 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
| 4 * | |
| 5 * This code is free software; you can redistribute it and/or modify it | |
| 6 * under the terms of the GNU General Public License version 2 only, as | |
| 7 * published by the Free Software Foundation. | |
| 8 * | |
| 9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
| 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
| 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
| 12 * version 2 for more details (a copy is included in the LICENSE file that | |
| 13 * accompanied this code). | |
| 14 * | |
| 15 * You should have received a copy of the GNU General Public License version | |
| 16 * 2 along with this work; if not, write to the Free Software Foundation, | |
| 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
| 18 * | |
|
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1295
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1295
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
|
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1295
diff
changeset
|
21 * questions. |
| 0 | 22 * |
| 23 */ | |
| 24 | |
| 1972 | 25 #include "precompiled.hpp" |
| 26 #include "c1/c1_MacroAssembler.hpp" | |
| 27 #include "c1/c1_Runtime1.hpp" | |
| 28 #include "classfile/systemDictionary.hpp" | |
| 29 #include "gc_interface/collectedHeap.hpp" | |
| 30 #include "interpreter/interpreter.hpp" | |
| 31 #include "oops/arrayOop.hpp" | |
| 32 #include "oops/markOop.hpp" | |
| 33 #include "runtime/basicLock.hpp" | |
| 34 #include "runtime/biasedLocking.hpp" | |
| 35 #include "runtime/os.hpp" | |
| 36 #include "runtime/stubRoutines.hpp" | |
| 0 | 37 |
| 38 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) { | |
| 39 Label L; | |
| 40 const Register temp_reg = G3_scratch; | |
| 41 // Note: needs more testing of out-of-line vs. inline slow case | |
| 42 verify_oop(receiver); | |
| 2002 | 43 load_klass(receiver, temp_reg); |
| 0 | 44 cmp(temp_reg, iCache); |
| 45 brx(Assembler::equal, true, Assembler::pt, L); | |
| 46 delayed()->nop(); | |
| 727 | 47 AddressLiteral ic_miss(SharedRuntime::get_ic_miss_stub()); |
| 48 jump_to(ic_miss, temp_reg); | |
| 0 | 49 delayed()->nop(); |
| 50 align(CodeEntryAlignment); | |
| 51 bind(L); | |
| 52 } | |
| 53 | |
| 54 | |
| 55 void C1_MacroAssembler::explicit_null_check(Register base) { | |
| 56 Unimplemented(); | |
| 57 } | |
| 58 | |
| 59 | |
| 60 void C1_MacroAssembler::build_frame(int frame_size_in_bytes) { | |
| 61 | |
| 62 generate_stack_overflow_check(frame_size_in_bytes); | |
| 63 // Create the frame. | |
| 64 save_frame_c1(frame_size_in_bytes); | |
| 65 } | |
| 66 | |
| 67 | |
| 68 void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) { | |
| 69 if (C1Breakpoint) breakpoint_trap(); | |
| 70 inline_cache_check(receiver, ic_klass); | |
| 71 } | |
| 72 | |
| 73 | |
| 74 void C1_MacroAssembler::verified_entry() { | |
| 75 if (C1Breakpoint) breakpoint_trap(); | |
| 76 // build frame | |
| 77 verify_FPU(0, "method_entry"); | |
| 78 } | |
| 79 | |
| 80 | |
| 81 void C1_MacroAssembler::lock_object(Register Rmark, Register Roop, Register Rbox, Register Rscratch, Label& slow_case) { | |
| 82 assert_different_registers(Rmark, Roop, Rbox, Rscratch); | |
| 83 | |
| 84 Label done; | |
| 85 | |
| 727 | 86 Address mark_addr(Roop, oopDesc::mark_offset_in_bytes()); |
| 0 | 87 |
| 88 // The following move must be the first instruction of emitted since debug | |
| 89 // information may be generated for it. | |
| 90 // Load object header | |
| 91 ld_ptr(mark_addr, Rmark); | |
| 92 | |
| 93 verify_oop(Roop); | |
| 94 | |
| 95 // save object being locked into the BasicObjectLock | |
| 96 st_ptr(Roop, Rbox, BasicObjectLock::obj_offset_in_bytes()); | |
| 97 | |
| 98 if (UseBiasedLocking) { | |
| 99 biased_locking_enter(Roop, Rmark, Rscratch, done, &slow_case); | |
| 100 } | |
| 101 | |
| 102 // Save Rbox in Rscratch to be used for the cas operation | |
| 103 mov(Rbox, Rscratch); | |
| 104 | |
| 105 // and mark it unlocked | |
| 106 or3(Rmark, markOopDesc::unlocked_value, Rmark); | |
| 107 | |
| 108 // save unlocked object header into the displaced header location on the stack | |
| 109 st_ptr(Rmark, Rbox, BasicLock::displaced_header_offset_in_bytes()); | |
| 110 | |
| 111 // compare object markOop with Rmark and if equal exchange Rscratch with object markOop | |
| 112 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
| 113 casx_under_lock(mark_addr.base(), Rmark, Rscratch, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
| 114 // if compare/exchange succeeded we found an unlocked object and we now have locked it | |
| 115 // hence we are done | |
| 116 cmp(Rmark, Rscratch); | |
| 117 brx(Assembler::equal, false, Assembler::pt, done); | |
| 118 delayed()->sub(Rscratch, SP, Rscratch); //pull next instruction into delay slot | |
| 119 // we did not find an unlocked object so see if this is a recursive case | |
| 120 // sub(Rscratch, SP, Rscratch); | |
| 121 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); | |
| 122 andcc(Rscratch, 0xfffff003, Rscratch); | |
| 123 brx(Assembler::notZero, false, Assembler::pn, slow_case); | |
| 124 delayed()->st_ptr(Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes()); | |
| 125 bind(done); | |
| 126 } | |
| 127 | |
| 128 | |
| 129 void C1_MacroAssembler::unlock_object(Register Rmark, Register Roop, Register Rbox, Label& slow_case) { | |
| 130 assert_different_registers(Rmark, Roop, Rbox); | |
| 131 | |
| 132 Label done; | |
| 133 | |
| 727 | 134 Address mark_addr(Roop, oopDesc::mark_offset_in_bytes()); |
| 0 | 135 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); |
| 136 | |
| 137 if (UseBiasedLocking) { | |
| 138 // load the object out of the BasicObjectLock | |
| 139 ld_ptr(Rbox, BasicObjectLock::obj_offset_in_bytes(), Roop); | |
| 140 verify_oop(Roop); | |
| 141 biased_locking_exit(mark_addr, Rmark, done); | |
| 142 } | |
| 143 // Test first it it is a fast recursive unlock | |
| 144 ld_ptr(Rbox, BasicLock::displaced_header_offset_in_bytes(), Rmark); | |
| 145 br_null(Rmark, false, Assembler::pt, done); | |
| 146 delayed()->nop(); | |
| 147 if (!UseBiasedLocking) { | |
| 148 // load object | |
| 149 ld_ptr(Rbox, BasicObjectLock::obj_offset_in_bytes(), Roop); | |
| 150 verify_oop(Roop); | |
| 151 } | |
| 152 | |
| 153 // Check if it is still a light weight lock, this is is true if we see | |
| 154 // the stack address of the basicLock in the markOop of the object | |
| 155 casx_under_lock(mark_addr.base(), Rbox, Rmark, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
| 156 cmp(Rbox, Rmark); | |
| 157 | |
| 158 brx(Assembler::notEqual, false, Assembler::pn, slow_case); | |
| 159 delayed()->nop(); | |
| 160 // Done | |
| 161 bind(done); | |
| 162 } | |
| 163 | |
| 164 | |
| 165 void C1_MacroAssembler::try_allocate( | |
| 166 Register obj, // result: pointer to object after successful allocation | |
| 167 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise | |
| 168 int con_size_in_bytes, // object size in bytes if known at compile time | |
|
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2002
diff
changeset
|
169 Register t1, // temp register, must be global register for incr_allocated_bytes |
| 0 | 170 Register t2, // temp register |
| 171 Label& slow_case // continuation point if fast allocation fails | |
| 172 ) { | |
| 173 if (UseTLAB) { | |
| 174 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case); | |
| 175 } else { | |
| 176 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case); | |
|
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2002
diff
changeset
|
177 incr_allocated_bytes(var_size_in_bytes, con_size_in_bytes, t1); |
| 0 | 178 } |
| 179 } | |
| 180 | |
| 181 | |
| 182 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) { | |
| 183 assert_different_registers(obj, klass, len, t1, t2); | |
| 184 if (UseBiasedLocking && !len->is_valid()) { | |
| 185 ld_ptr(klass, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes(), t1); | |
| 186 } else { | |
| 187 set((intx)markOopDesc::prototype(), t1); | |
| 188 } | |
| 2002 | 189 st_ptr(t1, obj, oopDesc::mark_offset_in_bytes()); |
| 190 if (UseCompressedOops) { | |
| 191 // Save klass | |
| 192 mov(klass, t1); | |
| 193 encode_heap_oop_not_null(t1); | |
| 194 stw(t1, obj, oopDesc::klass_offset_in_bytes()); | |
| 195 } else { | |
| 196 st_ptr(klass, obj, oopDesc::klass_offset_in_bytes()); | |
| 197 } | |
| 198 if (len->is_valid()) st(len, obj, arrayOopDesc::length_offset_in_bytes()); | |
| 199 else if (UseCompressedOops) { | |
| 200 store_klass_gap(G0, obj); | |
| 201 } | |
| 0 | 202 } |
| 203 | |
| 204 | |
| 205 void C1_MacroAssembler::initialize_body(Register base, Register index) { | |
| 206 assert_different_registers(base, index); | |
| 207 Label loop; | |
| 208 bind(loop); | |
| 209 subcc(index, HeapWordSize, index); | |
| 210 brx(Assembler::greaterEqual, true, Assembler::pt, loop); | |
| 211 delayed()->st_ptr(G0, base, index); | |
| 212 } | |
| 213 | |
| 214 | |
| 215 void C1_MacroAssembler::allocate_object( | |
| 216 Register obj, // result: pointer to object after successful allocation | |
| 217 Register t1, // temp register | |
|
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2002
diff
changeset
|
218 Register t2, // temp register, must be a global register for try_allocate |
| 0 | 219 Register t3, // temp register |
| 220 int hdr_size, // object header size in words | |
| 221 int obj_size, // object size in words | |
| 222 Register klass, // object klass | |
| 223 Label& slow_case // continuation point if fast allocation fails | |
| 224 ) { | |
| 225 assert_different_registers(obj, t1, t2, t3, klass); | |
| 226 assert(klass == G5, "must be G5"); | |
| 227 | |
| 228 // allocate space & initialize header | |
| 229 if (!is_simm13(obj_size * wordSize)) { | |
| 230 // would need to use extra register to load | |
| 231 // object size => go the slow case for now | |
| 232 br(Assembler::always, false, Assembler::pt, slow_case); | |
| 233 delayed()->nop(); | |
| 234 return; | |
| 235 } | |
| 236 try_allocate(obj, noreg, obj_size * wordSize, t2, t3, slow_case); | |
| 237 | |
| 238 initialize_object(obj, klass, noreg, obj_size * HeapWordSize, t1, t2); | |
| 239 } | |
| 240 | |
| 241 void C1_MacroAssembler::initialize_object( | |
| 242 Register obj, // result: pointer to object after successful allocation | |
| 243 Register klass, // object klass | |
| 244 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise | |
| 245 int con_size_in_bytes, // object size in bytes if known at compile time | |
| 246 Register t1, // temp register | |
| 247 Register t2 // temp register | |
| 248 ) { | |
| 2002 | 249 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize; |
| 0 | 250 |
| 251 initialize_header(obj, klass, noreg, t1, t2); | |
| 252 | |
| 253 #ifdef ASSERT | |
| 254 { | |
| 255 Label ok; | |
| 256 ld(klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), t1); | |
| 257 if (var_size_in_bytes != noreg) { | |
| 258 cmp(t1, var_size_in_bytes); | |
| 259 } else { | |
| 260 cmp(t1, con_size_in_bytes); | |
| 261 } | |
| 262 brx(Assembler::equal, false, Assembler::pt, ok); | |
| 263 delayed()->nop(); | |
| 264 stop("bad size in initialize_object"); | |
| 265 should_not_reach_here(); | |
| 266 | |
| 267 bind(ok); | |
| 268 } | |
| 269 | |
| 270 #endif | |
| 271 | |
| 272 // initialize body | |
| 273 const int threshold = 5 * HeapWordSize; // approximate break even point for code size | |
| 274 if (var_size_in_bytes != noreg) { | |
| 275 // use a loop | |
| 276 add(obj, hdr_size_in_bytes, t1); // compute address of first element | |
| 277 sub(var_size_in_bytes, hdr_size_in_bytes, t2); // compute size of body | |
| 278 initialize_body(t1, t2); | |
| 279 #ifndef _LP64 | |
| 280 } else if (VM_Version::v9_instructions_work() && con_size_in_bytes < threshold * 2) { | |
| 281 // on v9 we can do double word stores to fill twice as much space. | |
| 282 assert(hdr_size_in_bytes % 8 == 0, "double word aligned"); | |
| 283 assert(con_size_in_bytes % 8 == 0, "double word aligned"); | |
| 284 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += 2 * HeapWordSize) stx(G0, obj, i); | |
| 285 #endif | |
| 286 } else if (con_size_in_bytes <= threshold) { | |
| 287 // use explicit NULL stores | |
| 288 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += HeapWordSize) st_ptr(G0, obj, i); | |
| 289 } else if (con_size_in_bytes > hdr_size_in_bytes) { | |
| 290 // use a loop | |
| 291 const Register base = t1; | |
| 292 const Register index = t2; | |
| 293 add(obj, hdr_size_in_bytes, base); // compute address of first element | |
| 294 // compute index = number of words to clear | |
| 295 set(con_size_in_bytes - hdr_size_in_bytes, index); | |
| 296 initialize_body(base, index); | |
| 297 } | |
| 298 | |
|
780
c96bf21b756f
6788527: Server vm intermittently fails with assertion "live value must not be garbage" with fastdebug bits
kvn
parents:
727
diff
changeset
|
299 if (CURRENT_ENV->dtrace_alloc_probes()) { |
| 0 | 300 assert(obj == O0, "must be"); |
| 301 call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)), | |
| 302 relocInfo::runtime_call_type); | |
| 303 delayed()->nop(); | |
| 304 } | |
| 305 | |
| 306 verify_oop(obj); | |
| 307 } | |
| 308 | |
| 309 | |
| 310 void C1_MacroAssembler::allocate_array( | |
| 311 Register obj, // result: pointer to array after successful allocation | |
| 312 Register len, // array length | |
| 313 Register t1, // temp register | |
| 314 Register t2, // temp register | |
| 315 Register t3, // temp register | |
| 316 int hdr_size, // object header size in words | |
| 317 int elt_size, // element size in bytes | |
| 318 Register klass, // object klass | |
| 319 Label& slow_case // continuation point if fast allocation fails | |
| 320 ) { | |
| 321 assert_different_registers(obj, len, t1, t2, t3, klass); | |
| 322 assert(klass == G5, "must be G5"); | |
| 323 assert(t1 == G1, "must be G1"); | |
| 324 | |
| 325 // determine alignment mask | |
| 326 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work"); | |
| 327 | |
| 328 // check for negative or excessive length | |
| 329 // note: the maximum length allowed is chosen so that arrays of any | |
| 330 // element size with this length are always smaller or equal | |
| 331 // to the largest integer (i.e., array size computation will | |
| 332 // not overflow) | |
| 333 set(max_array_allocation_length, t1); | |
| 334 cmp(len, t1); | |
| 335 br(Assembler::greaterUnsigned, false, Assembler::pn, slow_case); | |
| 336 | |
| 337 // compute array size | |
| 338 // note: if 0 <= len <= max_length, len*elt_size + header + alignment is | |
| 339 // smaller or equal to the largest integer; also, since top is always | |
| 340 // aligned, we can do the alignment here instead of at the end address | |
| 341 // computation | |
| 342 const Register arr_size = t1; | |
| 343 switch (elt_size) { | |
| 344 case 1: delayed()->mov(len, arr_size); break; | |
| 345 case 2: delayed()->sll(len, 1, arr_size); break; | |
| 346 case 4: delayed()->sll(len, 2, arr_size); break; | |
| 347 case 8: delayed()->sll(len, 3, arr_size); break; | |
| 348 default: ShouldNotReachHere(); | |
| 349 } | |
| 350 add(arr_size, hdr_size * wordSize + MinObjAlignmentInBytesMask, arr_size); // add space for header & alignment | |
| 351 and3(arr_size, ~MinObjAlignmentInBytesMask, arr_size); // align array size | |
| 352 | |
| 353 // allocate space & initialize header | |
| 354 if (UseTLAB) { | |
| 355 tlab_allocate(obj, arr_size, 0, t2, slow_case); | |
| 356 } else { | |
| 357 eden_allocate(obj, arr_size, 0, t2, t3, slow_case); | |
| 358 } | |
| 359 initialize_header(obj, klass, len, t2, t3); | |
| 360 | |
| 361 // initialize body | |
| 362 const Register base = t2; | |
| 363 const Register index = t3; | |
| 364 add(obj, hdr_size * wordSize, base); // compute address of first element | |
| 365 sub(arr_size, hdr_size * wordSize, index); // compute index = number of words to clear | |
| 366 initialize_body(base, index); | |
| 367 | |
|
780
c96bf21b756f
6788527: Server vm intermittently fails with assertion "live value must not be garbage" with fastdebug bits
kvn
parents:
727
diff
changeset
|
368 if (CURRENT_ENV->dtrace_alloc_probes()) { |
| 0 | 369 assert(obj == O0, "must be"); |
| 370 call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)), | |
| 371 relocInfo::runtime_call_type); | |
| 372 delayed()->nop(); | |
| 373 } | |
| 374 | |
| 375 verify_oop(obj); | |
| 376 } | |
| 377 | |
| 378 | |
| 379 #ifndef PRODUCT | |
| 380 | |
| 381 void C1_MacroAssembler::verify_stack_oop(int stack_offset) { | |
| 382 if (!VerifyOops) return; | |
| 727 | 383 verify_oop_addr(Address(SP, stack_offset + STACK_BIAS)); |
| 0 | 384 } |
| 385 | |
| 386 void C1_MacroAssembler::verify_not_null_oop(Register r) { | |
| 387 Label not_null; | |
| 388 br_zero(Assembler::notEqual, false, Assembler::pt, r, not_null); | |
| 389 delayed()->nop(); | |
| 390 stop("non-null oop required"); | |
| 391 bind(not_null); | |
| 392 if (!VerifyOops) return; | |
| 393 verify_oop(r); | |
| 394 } | |
| 395 | |
| 396 void C1_MacroAssembler::invalidate_registers(bool iregisters, bool lregisters, bool oregisters, | |
| 397 Register preserve1, Register preserve2) { | |
| 398 if (iregisters) { | |
| 399 for (int i = 0; i < 6; i++) { | |
| 400 Register r = as_iRegister(i); | |
| 401 if (r != preserve1 && r != preserve2) set(0xdead, r); | |
| 402 } | |
| 403 } | |
| 404 if (oregisters) { | |
| 405 for (int i = 0; i < 6; i++) { | |
| 406 Register r = as_oRegister(i); | |
| 407 if (r != preserve1 && r != preserve2) set(0xdead, r); | |
| 408 } | |
| 409 } | |
| 410 if (lregisters) { | |
| 411 for (int i = 0; i < 8; i++) { | |
| 412 Register r = as_lRegister(i); | |
| 413 if (r != preserve1 && r != preserve2) set(0xdead, r); | |
| 414 } | |
| 415 } | |
| 416 } | |
| 417 | |
| 418 | |
| 419 #endif |