1 /* 2 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. 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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * 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 verify_oop(receiver); 33 load_klass(receiver, temp_reg); 34 cmp(temp_reg, iCache); 35 brx(Assembler::equal, true, Assembler::pt, L); 36 delayed()->nop(); 37 AddressLiteral ic_miss(SharedRuntime::get_ic_miss_stub()); 38 jump_to(ic_miss, temp_reg); 39 delayed()->nop(); 40 align(CodeEntryAlignment); 41 bind(L); 42 } 43 44 45 void C1_MacroAssembler::explicit_null_check(Register base) { 46 Unimplemented(); 47 } 48 49 50 void C1_MacroAssembler::build_frame(int frame_size_in_bytes) { 51 52 generate_stack_overflow_check(frame_size_in_bytes); 53 // Create the frame. 54 save_frame_c1(frame_size_in_bytes); 55 } 56 57 58 void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) { 59 if (C1Breakpoint) breakpoint_trap(); 60 inline_cache_check(receiver, ic_klass); 61 } 62 63 64 void C1_MacroAssembler::verified_entry() { 65 if (C1Breakpoint) breakpoint_trap(); 66 // build frame 67 verify_FPU(0, "method_entry"); 68 } 69 70 71 void C1_MacroAssembler::lock_object(Register Rmark, Register Roop, Register Rbox, Register Rscratch, Label& slow_case) { 72 assert_different_registers(Rmark, Roop, Rbox, Rscratch); 73 74 Label done; 75 76 Address mark_addr(Roop, oopDesc::mark_offset_in_bytes()); 77 78 // The following move must be the first instruction of emitted since debug 79 // information may be generated for it. 80 // Load object header 81 ld_ptr(mark_addr, Rmark); 82 83 verify_oop(Roop); 84 85 // save object being locked into the BasicObjectLock 86 st_ptr(Roop, Rbox, BasicObjectLock::obj_offset_in_bytes()); 87 88 if (UseBiasedLocking) { 89 biased_locking_enter(Roop, Rmark, Rscratch, done, &slow_case); 90 } 91 92 // Save Rbox in Rscratch to be used for the cas operation 93 mov(Rbox, Rscratch); 94 95 // and mark it unlocked 96 or3(Rmark, markOopDesc::unlocked_value, Rmark); 97 98 // save unlocked object header into the displaced header location on the stack 99 st_ptr(Rmark, Rbox, BasicLock::displaced_header_offset_in_bytes()); 100 101 // compare object markOop with Rmark and if equal exchange Rscratch with object markOop 102 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 103 casx_under_lock(mark_addr.base(), Rmark, Rscratch, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); 104 // if compare/exchange succeeded we found an unlocked object and we now have locked it 105 // hence we are done 106 cmp(Rmark, Rscratch); 107 brx(Assembler::equal, false, Assembler::pt, done); 108 delayed()->sub(Rscratch, SP, Rscratch); //pull next instruction into delay slot 109 // we did not find an unlocked object so see if this is a recursive case 110 // sub(Rscratch, SP, Rscratch); 111 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); 112 andcc(Rscratch, 0xfffff003, Rscratch); 113 brx(Assembler::notZero, false, Assembler::pn, slow_case); 114 delayed()->st_ptr(Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes()); 115 bind(done); 116 } 117 118 119 void C1_MacroAssembler::unlock_object(Register Rmark, Register Roop, Register Rbox, Label& slow_case) { 120 assert_different_registers(Rmark, Roop, Rbox); 121 122 Label done; 123 124 Address mark_addr(Roop, oopDesc::mark_offset_in_bytes()); 125 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); 126 127 if (UseBiasedLocking) { 128 // load the object out of the BasicObjectLock 129 ld_ptr(Rbox, BasicObjectLock::obj_offset_in_bytes(), Roop); 130 verify_oop(Roop); 131 biased_locking_exit(mark_addr, Rmark, done); 132 } 133 // Test first it it is a fast recursive unlock 134 ld_ptr(Rbox, BasicLock::displaced_header_offset_in_bytes(), Rmark); 135 br_null(Rmark, false, Assembler::pt, done); 136 delayed()->nop(); 137 if (!UseBiasedLocking) { 138 // load object 139 ld_ptr(Rbox, BasicObjectLock::obj_offset_in_bytes(), Roop); 140 verify_oop(Roop); 141 } 142 143 // Check if it is still a light weight lock, this is is true if we see 144 // the stack address of the basicLock in the markOop of the object 145 casx_under_lock(mark_addr.base(), Rbox, Rmark, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); 146 cmp(Rbox, Rmark); 147 148 brx(Assembler::notEqual, false, Assembler::pn, slow_case); 149 delayed()->nop(); 150 // Done 151 bind(done); 152 } 153 154 155 void C1_MacroAssembler::try_allocate( 156 Register obj, // result: pointer to object after successful allocation 157 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 158 int con_size_in_bytes, // object size in bytes if known at compile time 159 Register t1, // temp register 160 Register t2, // temp register 161 Label& slow_case // continuation point if fast allocation fails 162 ) { 163 if (UseTLAB) { 164 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case); 165 } else { 166 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case); 167 } 168 } 169 170 171 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) { 172 assert_different_registers(obj, klass, len, t1, t2); 173 if (UseBiasedLocking && !len->is_valid()) { 174 ld_ptr(klass, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes(), t1); 175 } else { 176 set((intx)markOopDesc::prototype(), t1); 177 } 178 st_ptr(t1, obj, oopDesc::mark_offset_in_bytes()); 179 #ifdef _LP64 180 if (UseCompressedOops) { 181 // Save klass 182 mov(klass, t1); 183 encode_heap_oop(t1); 184 stw(t1, obj, oopDesc::klass_offset_in_bytes()); 185 } else 186 #endif 187 st_ptr(klass, obj, oopDesc::klass_offset_in_bytes()); 188 if (len->is_valid()) st(len, obj, arrayOopDesc::length_offset_in_bytes()); 189 #ifdef _LP64 190 else if (UseCompressedOops) { 191 store_klass_gap(G0, obj); 192 } 193 #endif 194 } 195 196 197 void C1_MacroAssembler::initialize_body(Register base, Register index) { 198 assert_different_registers(base, index); 199 Label loop; 200 bind(loop); 201 subcc(index, HeapWordSize, index); 202 brx(Assembler::greaterEqual, true, Assembler::pt, loop); 203 delayed()->st_ptr(G0, base, index); 204 } 205 206 207 void C1_MacroAssembler::allocate_object( 208 Register obj, // result: pointer to object after successful allocation 209 Register t1, // temp register 210 Register t2, // temp register 211 Register t3, // temp register 212 int hdr_size, // object header size in words 213 int obj_size, // object size in words 214 Register klass, // object klass 215 Label& slow_case // continuation point if fast allocation fails 216 ) { 217 assert_different_registers(obj, t1, t2, t3, klass); 218 assert(klass == G5, "must be G5"); 219 220 // allocate space & initialize header 221 if (!is_simm13(obj_size * wordSize)) { 222 // would need to use extra register to load 223 // object size => go the slow case for now 224 br(Assembler::always, false, Assembler::pt, slow_case); 225 delayed()->nop(); 226 return; 227 } 228 try_allocate(obj, noreg, obj_size * wordSize, t2, t3, slow_case); 229 230 initialize_object(obj, klass, noreg, obj_size * HeapWordSize, t1, t2); 231 } 232 233 void C1_MacroAssembler::initialize_object( 234 Register obj, // result: pointer to object after successful allocation 235 Register klass, // object klass 236 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 237 int con_size_in_bytes, // object size in bytes if known at compile time 238 Register t1, // temp register 239 Register t2 // temp register 240 ) { 241 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize; 242 243 initialize_header(obj, klass, noreg, t1, t2); 244 245 #ifdef ASSERT 246 { 247 Label ok; 248 ld(klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), t1); 249 if (var_size_in_bytes != noreg) { 250 cmp(t1, var_size_in_bytes); 251 } else { 252 cmp(t1, con_size_in_bytes); 253 } 254 brx(Assembler::equal, false, Assembler::pt, ok); 255 delayed()->nop(); 256 stop("bad size in initialize_object"); 257 should_not_reach_here(); 258 259 bind(ok); 260 } 261 262 #endif 263 264 // initialize body 265 const int threshold = 5 * HeapWordSize; // approximate break even point for code size 266 if (var_size_in_bytes != noreg) { 267 // use a loop 268 add(obj, hdr_size_in_bytes, t1); // compute address of first element 269 sub(var_size_in_bytes, hdr_size_in_bytes, t2); // compute size of body 270 initialize_body(t1, t2); 271 #ifndef _LP64 272 } else if (VM_Version::v9_instructions_work() && con_size_in_bytes < threshold * 2) { 273 // on v9 we can do double word stores to fill twice as much space. 274 assert(hdr_size_in_bytes % 8 == 0, "double word aligned"); 275 assert(con_size_in_bytes % 8 == 0, "double word aligned"); 276 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += 2 * HeapWordSize) stx(G0, obj, i); 277 #endif 278 } else if (con_size_in_bytes <= threshold) { 279 // use explicit NULL stores 280 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += HeapWordSize) st_ptr(G0, obj, i); 281 } else if (con_size_in_bytes > hdr_size_in_bytes) { 282 // use a loop 283 const Register base = t1; 284 const Register index = t2; 285 add(obj, hdr_size_in_bytes, base); // compute address of first element 286 // compute index = number of words to clear 287 set(con_size_in_bytes - hdr_size_in_bytes, index); 288 initialize_body(base, index); 289 } 290 291 if (CURRENT_ENV->dtrace_alloc_probes()) { 292 assert(obj == O0, "must be"); 293 call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)), 294 relocInfo::runtime_call_type); 295 delayed()->nop(); 296 } 297 298 verify_oop(obj); 299 } 300 301 302 void C1_MacroAssembler::allocate_array( 303 Register obj, // result: pointer to array after successful allocation 304 Register len, // array length 305 Register t1, // temp register 306 Register t2, // temp register 307 Register t3, // temp register 308 int hdr_size, // object header size in words 309 int elt_size, // element size in bytes 310 Register klass, // object klass 311 Label& slow_case // continuation point if fast allocation fails 312 ) { 313 assert_different_registers(obj, len, t1, t2, t3, klass); 314 assert(klass == G5, "must be G5"); 315 assert(t1 == G1, "must be G1"); 316 317 // determine alignment mask 318 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work"); 319 320 // check for negative or excessive length 321 // note: the maximum length allowed is chosen so that arrays of any 322 // element size with this length are always smaller or equal 323 // to the largest integer (i.e., array size computation will 324 // not overflow) 325 set(max_array_allocation_length, t1); 326 cmp(len, t1); 327 br(Assembler::greaterUnsigned, false, Assembler::pn, slow_case); 328 329 // compute array size 330 // note: if 0 <= len <= max_length, len*elt_size + header + alignment is 331 // smaller or equal to the largest integer; also, since top is always 332 // aligned, we can do the alignment here instead of at the end address 333 // computation 334 const Register arr_size = t1; 335 switch (elt_size) { 336 case 1: delayed()->mov(len, arr_size); break; 337 case 2: delayed()->sll(len, 1, arr_size); break; 338 case 4: delayed()->sll(len, 2, arr_size); break; 339 case 8: delayed()->sll(len, 3, arr_size); break; 340 default: ShouldNotReachHere(); 341 } 342 add(arr_size, hdr_size * wordSize + MinObjAlignmentInBytesMask, arr_size); // add space for header & alignment 343 and3(arr_size, ~MinObjAlignmentInBytesMask, arr_size); // align array size 344 345 // allocate space & initialize header 346 if (UseTLAB) { 347 tlab_allocate(obj, arr_size, 0, t2, slow_case); 348 } else { 349 eden_allocate(obj, arr_size, 0, t2, t3, slow_case); 350 } 351 initialize_header(obj, klass, len, t2, t3); 352 353 // initialize body 354 const Register base = t2; 355 const Register index = t3; 356 add(obj, hdr_size * wordSize, base); // compute address of first element 357 sub(arr_size, hdr_size * wordSize, index); // compute index = number of words to clear 358 initialize_body(base, index); 359 360 if (CURRENT_ENV->dtrace_alloc_probes()) { 361 assert(obj == O0, "must be"); 362 call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)), 363 relocInfo::runtime_call_type); 364 delayed()->nop(); 365 } 366 367 verify_oop(obj); 368 } 369 370 371 #ifndef PRODUCT 372 373 void C1_MacroAssembler::verify_stack_oop(int stack_offset) { 374 if (!VerifyOops) return; 375 verify_oop_addr(Address(SP, stack_offset + STACK_BIAS)); 376 } 377 378 void C1_MacroAssembler::verify_not_null_oop(Register r) { 379 Label not_null; 380 br_zero(Assembler::notEqual, false, Assembler::pt, r, not_null); 381 delayed()->nop(); 382 stop("non-null oop required"); 383 bind(not_null); 384 if (!VerifyOops) return; 385 verify_oop(r); 386 } 387 388 void C1_MacroAssembler::invalidate_registers(bool iregisters, bool lregisters, bool oregisters, 389 Register preserve1, Register preserve2) { 390 if (iregisters) { 391 for (int i = 0; i < 6; i++) { 392 Register r = as_iRegister(i); 393 if (r != preserve1 && r != preserve2) set(0xdead, r); 394 } 395 } 396 if (oregisters) { 397 for (int i = 0; i < 6; i++) { 398 Register r = as_oRegister(i); 399 if (r != preserve1 && r != preserve2) set(0xdead, r); 400 } 401 } 402 if (lregisters) { 403 for (int i = 0; i < 8; i++) { 404 Register r = as_lRegister(i); 405 if (r != preserve1 && r != preserve2) set(0xdead, r); 406 } 407 } 408 } 409 410 411 #endif