Mercurial > hg > truffle
annotate src/cpu/sparc/vm/c1_MacroAssembler_sparc.cpp @ 113:ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
Summary: Compressed oops in instances, arrays, and headers. Code contributors are coleenp, phh, never, swamyv
Reviewed-by: jmasa, kamg, acorn, tbell, kvn, rasbold
| author | coleenp |
|---|---|
| date | Sun, 13 Apr 2008 17:43:42 -0400 |
| parents | a61af66fc99e |
| children | d1605aabd0a1 |
| rev | line source |
|---|---|
| 0 | 1 /* |
| 2 * Copyright 1999-2007 Sun Microsystems, Inc. All Rights Reserved. | |
| 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 * | |
| 19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
| 20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
| 21 * have any questions. | |
| 22 * | |
| 23 */ | |
| 24 | |
| 25 #include "incls/_precompiled.incl" | |
| 26 #include "incls/_c1_MacroAssembler_sparc.cpp.incl" | |
| 27 | |
| 28 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) { | |
| 29 Label L; | |
| 30 const Register temp_reg = G3_scratch; | |
| 31 // Note: needs more testing of out-of-line vs. inline slow case | |
| 32 Address ic_miss(temp_reg, SharedRuntime::get_ic_miss_stub()); | |
| 33 verify_oop(receiver); | |
| 34 ld_ptr(receiver, oopDesc::klass_offset_in_bytes(), temp_reg); | |
| 35 cmp(temp_reg, iCache); | |
| 36 brx(Assembler::equal, true, Assembler::pt, L); | |
| 37 delayed()->nop(); | |
| 38 jump_to(ic_miss, 0); | |
| 39 delayed()->nop(); | |
| 40 align(CodeEntryAlignment); | |
| 41 bind(L); | |
| 42 } | |
| 43 | |
| 44 | |
| 45 void C1_MacroAssembler::method_exit(bool restore_frame) { | |
| 46 // this code must be structured this way so that the return | |
| 47 // instruction can be a safepoint. | |
| 48 if (restore_frame) { | |
| 49 restore(); | |
| 50 } | |
| 51 retl(); | |
| 52 delayed()->nop(); | |
| 53 } | |
| 54 | |
| 55 | |
| 56 void C1_MacroAssembler::explicit_null_check(Register base) { | |
| 57 Unimplemented(); | |
| 58 } | |
| 59 | |
| 60 | |
| 61 void C1_MacroAssembler::build_frame(int frame_size_in_bytes) { | |
| 62 | |
| 63 generate_stack_overflow_check(frame_size_in_bytes); | |
| 64 // Create the frame. | |
| 65 save_frame_c1(frame_size_in_bytes); | |
| 66 } | |
| 67 | |
| 68 | |
| 69 void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) { | |
| 70 if (C1Breakpoint) breakpoint_trap(); | |
| 71 inline_cache_check(receiver, ic_klass); | |
| 72 } | |
| 73 | |
| 74 | |
| 75 void C1_MacroAssembler::verified_entry() { | |
| 76 if (C1Breakpoint) breakpoint_trap(); | |
| 77 // build frame | |
| 78 verify_FPU(0, "method_entry"); | |
| 79 } | |
| 80 | |
| 81 | |
| 82 void C1_MacroAssembler::lock_object(Register Rmark, Register Roop, Register Rbox, Register Rscratch, Label& slow_case) { | |
| 83 assert_different_registers(Rmark, Roop, Rbox, Rscratch); | |
| 84 | |
| 85 Label done; | |
| 86 | |
| 87 Address mark_addr(Roop, 0, oopDesc::mark_offset_in_bytes()); | |
| 88 | |
| 89 // The following move must be the first instruction of emitted since debug | |
| 90 // information may be generated for it. | |
| 91 // Load object header | |
| 92 ld_ptr(mark_addr, Rmark); | |
| 93 | |
| 94 verify_oop(Roop); | |
| 95 | |
| 96 // save object being locked into the BasicObjectLock | |
| 97 st_ptr(Roop, Rbox, BasicObjectLock::obj_offset_in_bytes()); | |
| 98 | |
| 99 if (UseBiasedLocking) { | |
| 100 biased_locking_enter(Roop, Rmark, Rscratch, done, &slow_case); | |
| 101 } | |
| 102 | |
| 103 // Save Rbox in Rscratch to be used for the cas operation | |
| 104 mov(Rbox, Rscratch); | |
| 105 | |
| 106 // and mark it unlocked | |
| 107 or3(Rmark, markOopDesc::unlocked_value, Rmark); | |
| 108 | |
| 109 // save unlocked object header into the displaced header location on the stack | |
| 110 st_ptr(Rmark, Rbox, BasicLock::displaced_header_offset_in_bytes()); | |
| 111 | |
| 112 // compare object markOop with Rmark and if equal exchange Rscratch with object markOop | |
| 113 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
| 114 casx_under_lock(mark_addr.base(), Rmark, Rscratch, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
| 115 // if compare/exchange succeeded we found an unlocked object and we now have locked it | |
| 116 // hence we are done | |
| 117 cmp(Rmark, Rscratch); | |
| 118 brx(Assembler::equal, false, Assembler::pt, done); | |
| 119 delayed()->sub(Rscratch, SP, Rscratch); //pull next instruction into delay slot | |
| 120 // we did not find an unlocked object so see if this is a recursive case | |
| 121 // sub(Rscratch, SP, Rscratch); | |
| 122 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); | |
| 123 andcc(Rscratch, 0xfffff003, Rscratch); | |
| 124 brx(Assembler::notZero, false, Assembler::pn, slow_case); | |
| 125 delayed()->st_ptr(Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes()); | |
| 126 bind(done); | |
| 127 } | |
| 128 | |
| 129 | |
| 130 void C1_MacroAssembler::unlock_object(Register Rmark, Register Roop, Register Rbox, Label& slow_case) { | |
| 131 assert_different_registers(Rmark, Roop, Rbox); | |
| 132 | |
| 133 Label done; | |
| 134 | |
| 135 Address mark_addr(Roop, 0, oopDesc::mark_offset_in_bytes()); | |
| 136 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
| 137 | |
| 138 if (UseBiasedLocking) { | |
| 139 // load the object out of the BasicObjectLock | |
| 140 ld_ptr(Rbox, BasicObjectLock::obj_offset_in_bytes(), Roop); | |
| 141 verify_oop(Roop); | |
| 142 biased_locking_exit(mark_addr, Rmark, done); | |
| 143 } | |
| 144 // Test first it it is a fast recursive unlock | |
| 145 ld_ptr(Rbox, BasicLock::displaced_header_offset_in_bytes(), Rmark); | |
| 146 br_null(Rmark, false, Assembler::pt, done); | |
| 147 delayed()->nop(); | |
| 148 if (!UseBiasedLocking) { | |
| 149 // load object | |
| 150 ld_ptr(Rbox, BasicObjectLock::obj_offset_in_bytes(), Roop); | |
| 151 verify_oop(Roop); | |
| 152 } | |
| 153 | |
| 154 // Check if it is still a light weight lock, this is is true if we see | |
| 155 // the stack address of the basicLock in the markOop of the object | |
| 156 casx_under_lock(mark_addr.base(), Rbox, Rmark, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
| 157 cmp(Rbox, Rmark); | |
| 158 | |
| 159 brx(Assembler::notEqual, false, Assembler::pn, slow_case); | |
| 160 delayed()->nop(); | |
| 161 // Done | |
| 162 bind(done); | |
| 163 } | |
| 164 | |
| 165 | |
| 166 void C1_MacroAssembler::try_allocate( | |
| 167 Register obj, // result: pointer to object after successful allocation | |
| 168 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise | |
| 169 int con_size_in_bytes, // object size in bytes if known at compile time | |
| 170 Register t1, // temp register | |
| 171 Register t2, // temp register | |
| 172 Label& slow_case // continuation point if fast allocation fails | |
| 173 ) { | |
| 174 if (UseTLAB) { | |
| 175 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case); | |
| 176 } else { | |
| 177 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case); | |
| 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 } | |
| 189 st_ptr(t1 , obj, oopDesc::mark_offset_in_bytes ()); | |
| 190 st_ptr(klass, obj, oopDesc::klass_offset_in_bytes ()); | |
| 191 if (len->is_valid()) st(len , obj, arrayOopDesc::length_offset_in_bytes()); | |
| 192 } | |
| 193 | |
| 194 | |
| 195 void C1_MacroAssembler::initialize_body(Register base, Register index) { | |
| 196 assert_different_registers(base, index); | |
| 197 Label loop; | |
| 198 bind(loop); | |
| 199 subcc(index, HeapWordSize, index); | |
| 200 brx(Assembler::greaterEqual, true, Assembler::pt, loop); | |
| 201 delayed()->st_ptr(G0, base, index); | |
| 202 } | |
| 203 | |
| 204 | |
| 205 void C1_MacroAssembler::allocate_object( | |
| 206 Register obj, // result: pointer to object after successful allocation | |
| 207 Register t1, // temp register | |
| 208 Register t2, // temp register | |
| 209 Register t3, // temp register | |
| 210 int hdr_size, // object header size in words | |
| 211 int obj_size, // object size in words | |
| 212 Register klass, // object klass | |
| 213 Label& slow_case // continuation point if fast allocation fails | |
| 214 ) { | |
| 215 assert_different_registers(obj, t1, t2, t3, klass); | |
| 216 assert(klass == G5, "must be G5"); | |
| 217 | |
| 218 // allocate space & initialize header | |
| 219 if (!is_simm13(obj_size * wordSize)) { | |
| 220 // would need to use extra register to load | |
| 221 // object size => go the slow case for now | |
| 222 br(Assembler::always, false, Assembler::pt, slow_case); | |
| 223 delayed()->nop(); | |
| 224 return; | |
| 225 } | |
| 226 try_allocate(obj, noreg, obj_size * wordSize, t2, t3, slow_case); | |
| 227 | |
| 228 initialize_object(obj, klass, noreg, obj_size * HeapWordSize, t1, t2); | |
| 229 } | |
| 230 | |
| 231 void C1_MacroAssembler::initialize_object( | |
| 232 Register obj, // result: pointer to object after successful allocation | |
| 233 Register klass, // object klass | |
| 234 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise | |
| 235 int con_size_in_bytes, // object size in bytes if known at compile time | |
| 236 Register t1, // temp register | |
| 237 Register t2 // temp register | |
| 238 ) { | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
239 const int hdr_size_in_bytes = instanceOopDesc::base_offset_in_bytes(); |
| 0 | 240 |
| 241 initialize_header(obj, klass, noreg, t1, t2); | |
| 242 | |
| 243 #ifdef ASSERT | |
| 244 { | |
| 245 Label ok; | |
| 246 ld(klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), t1); | |
| 247 if (var_size_in_bytes != noreg) { | |
| 248 cmp(t1, var_size_in_bytes); | |
| 249 } else { | |
| 250 cmp(t1, con_size_in_bytes); | |
| 251 } | |
| 252 brx(Assembler::equal, false, Assembler::pt, ok); | |
| 253 delayed()->nop(); | |
| 254 stop("bad size in initialize_object"); | |
| 255 should_not_reach_here(); | |
| 256 | |
| 257 bind(ok); | |
| 258 } | |
| 259 | |
| 260 #endif | |
| 261 | |
| 262 // initialize body | |
| 263 const int threshold = 5 * HeapWordSize; // approximate break even point for code size | |
| 264 if (var_size_in_bytes != noreg) { | |
| 265 // use a loop | |
| 266 add(obj, hdr_size_in_bytes, t1); // compute address of first element | |
| 267 sub(var_size_in_bytes, hdr_size_in_bytes, t2); // compute size of body | |
| 268 initialize_body(t1, t2); | |
| 269 #ifndef _LP64 | |
| 270 } else if (VM_Version::v9_instructions_work() && con_size_in_bytes < threshold * 2) { | |
| 271 // on v9 we can do double word stores to fill twice as much space. | |
| 272 assert(hdr_size_in_bytes % 8 == 0, "double word aligned"); | |
| 273 assert(con_size_in_bytes % 8 == 0, "double word aligned"); | |
| 274 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += 2 * HeapWordSize) stx(G0, obj, i); | |
| 275 #endif | |
| 276 } else if (con_size_in_bytes <= threshold) { | |
| 277 // use explicit NULL stores | |
| 278 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += HeapWordSize) st_ptr(G0, obj, i); | |
| 279 } else if (con_size_in_bytes > hdr_size_in_bytes) { | |
| 280 // use a loop | |
| 281 const Register base = t1; | |
| 282 const Register index = t2; | |
| 283 add(obj, hdr_size_in_bytes, base); // compute address of first element | |
| 284 // compute index = number of words to clear | |
| 285 set(con_size_in_bytes - hdr_size_in_bytes, index); | |
| 286 initialize_body(base, index); | |
| 287 } | |
| 288 | |
| 289 if (DTraceAllocProbes) { | |
| 290 assert(obj == O0, "must be"); | |
| 291 call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)), | |
| 292 relocInfo::runtime_call_type); | |
| 293 delayed()->nop(); | |
| 294 } | |
| 295 | |
| 296 verify_oop(obj); | |
| 297 } | |
| 298 | |
| 299 | |
| 300 void C1_MacroAssembler::allocate_array( | |
| 301 Register obj, // result: pointer to array after successful allocation | |
| 302 Register len, // array length | |
| 303 Register t1, // temp register | |
| 304 Register t2, // temp register | |
| 305 Register t3, // temp register | |
| 306 int hdr_size, // object header size in words | |
| 307 int elt_size, // element size in bytes | |
| 308 Register klass, // object klass | |
| 309 Label& slow_case // continuation point if fast allocation fails | |
| 310 ) { | |
| 311 assert_different_registers(obj, len, t1, t2, t3, klass); | |
| 312 assert(klass == G5, "must be G5"); | |
| 313 assert(t1 == G1, "must be G1"); | |
| 314 | |
| 315 // determine alignment mask | |
| 316 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work"); | |
| 317 | |
| 318 // check for negative or excessive length | |
| 319 // note: the maximum length allowed is chosen so that arrays of any | |
| 320 // element size with this length are always smaller or equal | |
| 321 // to the largest integer (i.e., array size computation will | |
| 322 // not overflow) | |
| 323 set(max_array_allocation_length, t1); | |
| 324 cmp(len, t1); | |
| 325 br(Assembler::greaterUnsigned, false, Assembler::pn, slow_case); | |
| 326 | |
| 327 // compute array size | |
| 328 // note: if 0 <= len <= max_length, len*elt_size + header + alignment is | |
| 329 // smaller or equal to the largest integer; also, since top is always | |
| 330 // aligned, we can do the alignment here instead of at the end address | |
| 331 // computation | |
| 332 const Register arr_size = t1; | |
| 333 switch (elt_size) { | |
| 334 case 1: delayed()->mov(len, arr_size); break; | |
| 335 case 2: delayed()->sll(len, 1, arr_size); break; | |
| 336 case 4: delayed()->sll(len, 2, arr_size); break; | |
| 337 case 8: delayed()->sll(len, 3, arr_size); break; | |
| 338 default: ShouldNotReachHere(); | |
| 339 } | |
| 340 add(arr_size, hdr_size * wordSize + MinObjAlignmentInBytesMask, arr_size); // add space for header & alignment | |
| 341 and3(arr_size, ~MinObjAlignmentInBytesMask, arr_size); // align array size | |
| 342 | |
| 343 // allocate space & initialize header | |
| 344 if (UseTLAB) { | |
| 345 tlab_allocate(obj, arr_size, 0, t2, slow_case); | |
| 346 } else { | |
| 347 eden_allocate(obj, arr_size, 0, t2, t3, slow_case); | |
| 348 } | |
| 349 initialize_header(obj, klass, len, t2, t3); | |
| 350 | |
| 351 // initialize body | |
| 352 const Register base = t2; | |
| 353 const Register index = t3; | |
| 354 add(obj, hdr_size * wordSize, base); // compute address of first element | |
| 355 sub(arr_size, hdr_size * wordSize, index); // compute index = number of words to clear | |
| 356 initialize_body(base, index); | |
| 357 | |
| 358 if (DTraceAllocProbes) { | |
| 359 assert(obj == O0, "must be"); | |
| 360 call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)), | |
| 361 relocInfo::runtime_call_type); | |
| 362 delayed()->nop(); | |
| 363 } | |
| 364 | |
| 365 verify_oop(obj); | |
| 366 } | |
| 367 | |
| 368 | |
| 369 #ifndef PRODUCT | |
| 370 | |
| 371 void C1_MacroAssembler::verify_stack_oop(int stack_offset) { | |
| 372 if (!VerifyOops) return; | |
| 373 verify_oop_addr(Address(SP, 0, stack_offset + STACK_BIAS)); | |
| 374 } | |
| 375 | |
| 376 void C1_MacroAssembler::verify_not_null_oop(Register r) { | |
| 377 Label not_null; | |
| 378 br_zero(Assembler::notEqual, false, Assembler::pt, r, not_null); | |
| 379 delayed()->nop(); | |
| 380 stop("non-null oop required"); | |
| 381 bind(not_null); | |
| 382 if (!VerifyOops) return; | |
| 383 verify_oop(r); | |
| 384 } | |
| 385 | |
| 386 void C1_MacroAssembler::invalidate_registers(bool iregisters, bool lregisters, bool oregisters, | |
| 387 Register preserve1, Register preserve2) { | |
| 388 if (iregisters) { | |
| 389 for (int i = 0; i < 6; i++) { | |
| 390 Register r = as_iRegister(i); | |
| 391 if (r != preserve1 && r != preserve2) set(0xdead, r); | |
| 392 } | |
| 393 } | |
| 394 if (oregisters) { | |
| 395 for (int i = 0; i < 6; i++) { | |
| 396 Register r = as_oRegister(i); | |
| 397 if (r != preserve1 && r != preserve2) set(0xdead, r); | |
| 398 } | |
| 399 } | |
| 400 if (lregisters) { | |
| 401 for (int i = 0; i < 8; i++) { | |
| 402 Register r = as_lRegister(i); | |
| 403 if (r != preserve1 && r != preserve2) set(0xdead, r); | |
| 404 } | |
| 405 } | |
| 406 } | |
| 407 | |
| 408 | |
| 409 #endif |