1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000 1.2 +++ b/src/cpu/x86/vm/templateTable_x86_32.cpp Wed Apr 27 01:25:04 2016 +0800 1.3 @@ -0,0 +1,3689 @@ 1.4 +/* 1.5 + * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. 1.6 + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 1.7 + * 1.8 + * This code is free software; you can redistribute it and/or modify it 1.9 + * under the terms of the GNU General Public License version 2 only, as 1.10 + * published by the Free Software Foundation. 1.11 + * 1.12 + * This code is distributed in the hope that it will be useful, but WITHOUT 1.13 + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 1.14 + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 1.15 + * version 2 for more details (a copy is included in the LICENSE file that 1.16 + * accompanied this code). 1.17 + * 1.18 + * You should have received a copy of the GNU General Public License version 1.19 + * 2 along with this work; if not, write to the Free Software Foundation, 1.20 + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 1.21 + * 1.22 + * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 1.23 + * or visit www.oracle.com if you need additional information or have any 1.24 + * questions. 1.25 + * 1.26 + */ 1.27 + 1.28 +#include "precompiled.hpp" 1.29 +#include "asm/macroAssembler.hpp" 1.30 +#include "interpreter/interpreter.hpp" 1.31 +#include "interpreter/interpreterRuntime.hpp" 1.32 +#include "interpreter/templateTable.hpp" 1.33 +#include "memory/universe.inline.hpp" 1.34 +#include "oops/methodData.hpp" 1.35 +#include "oops/objArrayKlass.hpp" 1.36 +#include "oops/oop.inline.hpp" 1.37 +#include "prims/methodHandles.hpp" 1.38 +#include "runtime/sharedRuntime.hpp" 1.39 +#include "runtime/stubRoutines.hpp" 1.40 +#include "runtime/synchronizer.hpp" 1.41 +#include "utilities/macros.hpp" 1.42 + 1.43 +#ifndef CC_INTERP 1.44 +#define __ _masm-> 1.45 + 1.46 +//---------------------------------------------------------------------------------------------------- 1.47 +// Platform-dependent initialization 1.48 + 1.49 +void TemplateTable::pd_initialize() { 1.50 + // No i486 specific initialization 1.51 +} 1.52 + 1.53 +//---------------------------------------------------------------------------------------------------- 1.54 +// Address computation 1.55 + 1.56 +// local variables 1.57 +static inline Address iaddress(int n) { 1.58 + return Address(rdi, Interpreter::local_offset_in_bytes(n)); 1.59 +} 1.60 + 1.61 +static inline Address laddress(int n) { return iaddress(n + 1); } 1.62 +static inline Address haddress(int n) { return iaddress(n + 0); } 1.63 +static inline Address faddress(int n) { return iaddress(n); } 1.64 +static inline Address daddress(int n) { return laddress(n); } 1.65 +static inline Address aaddress(int n) { return iaddress(n); } 1.66 + 1.67 +static inline Address iaddress(Register r) { 1.68 + return Address(rdi, r, Interpreter::stackElementScale()); 1.69 +} 1.70 +static inline Address laddress(Register r) { 1.71 + return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(1)); 1.72 +} 1.73 +static inline Address haddress(Register r) { 1.74 + return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0)); 1.75 +} 1.76 + 1.77 +static inline Address faddress(Register r) { return iaddress(r); } 1.78 +static inline Address daddress(Register r) { return laddress(r); } 1.79 +static inline Address aaddress(Register r) { return iaddress(r); } 1.80 + 1.81 +// expression stack 1.82 +// (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store 1.83 +// data beyond the rsp which is potentially unsafe in an MT environment; 1.84 +// an interrupt may overwrite that data.) 1.85 +static inline Address at_rsp () { 1.86 + return Address(rsp, 0); 1.87 +} 1.88 + 1.89 +// At top of Java expression stack which may be different than rsp(). It 1.90 +// isn't for category 1 objects. 1.91 +static inline Address at_tos () { 1.92 + Address tos = Address(rsp, Interpreter::expr_offset_in_bytes(0)); 1.93 + return tos; 1.94 +} 1.95 + 1.96 +static inline Address at_tos_p1() { 1.97 + return Address(rsp, Interpreter::expr_offset_in_bytes(1)); 1.98 +} 1.99 + 1.100 +static inline Address at_tos_p2() { 1.101 + return Address(rsp, Interpreter::expr_offset_in_bytes(2)); 1.102 +} 1.103 + 1.104 +// Condition conversion 1.105 +static Assembler::Condition j_not(TemplateTable::Condition cc) { 1.106 + switch (cc) { 1.107 + case TemplateTable::equal : return Assembler::notEqual; 1.108 + case TemplateTable::not_equal : return Assembler::equal; 1.109 + case TemplateTable::less : return Assembler::greaterEqual; 1.110 + case TemplateTable::less_equal : return Assembler::greater; 1.111 + case TemplateTable::greater : return Assembler::lessEqual; 1.112 + case TemplateTable::greater_equal: return Assembler::less; 1.113 + } 1.114 + ShouldNotReachHere(); 1.115 + return Assembler::zero; 1.116 +} 1.117 + 1.118 + 1.119 +//---------------------------------------------------------------------------------------------------- 1.120 +// Miscelaneous helper routines 1.121 + 1.122 +// Store an oop (or NULL) at the address described by obj. 1.123 +// If val == noreg this means store a NULL 1.124 + 1.125 +static void do_oop_store(InterpreterMacroAssembler* _masm, 1.126 + Address obj, 1.127 + Register val, 1.128 + BarrierSet::Name barrier, 1.129 + bool precise) { 1.130 + assert(val == noreg || val == rax, "parameter is just for looks"); 1.131 + switch (barrier) { 1.132 +#if INCLUDE_ALL_GCS 1.133 + case BarrierSet::G1SATBCT: 1.134 + case BarrierSet::G1SATBCTLogging: 1.135 + { 1.136 + // flatten object address if needed 1.137 + // We do it regardless of precise because we need the registers 1.138 + if (obj.index() == noreg && obj.disp() == 0) { 1.139 + if (obj.base() != rdx) { 1.140 + __ movl(rdx, obj.base()); 1.141 + } 1.142 + } else { 1.143 + __ leal(rdx, obj); 1.144 + } 1.145 + __ get_thread(rcx); 1.146 + __ save_bcp(); 1.147 + __ g1_write_barrier_pre(rdx /* obj */, 1.148 + rbx /* pre_val */, 1.149 + rcx /* thread */, 1.150 + rsi /* tmp */, 1.151 + val != noreg /* tosca_live */, 1.152 + false /* expand_call */); 1.153 + 1.154 + // Do the actual store 1.155 + // noreg means NULL 1.156 + if (val == noreg) { 1.157 + __ movptr(Address(rdx, 0), NULL_WORD); 1.158 + // No post barrier for NULL 1.159 + } else { 1.160 + __ movl(Address(rdx, 0), val); 1.161 + __ g1_write_barrier_post(rdx /* store_adr */, 1.162 + val /* new_val */, 1.163 + rcx /* thread */, 1.164 + rbx /* tmp */, 1.165 + rsi /* tmp2 */); 1.166 + } 1.167 + __ restore_bcp(); 1.168 + 1.169 + } 1.170 + break; 1.171 +#endif // INCLUDE_ALL_GCS 1.172 + case BarrierSet::CardTableModRef: 1.173 + case BarrierSet::CardTableExtension: 1.174 + { 1.175 + if (val == noreg) { 1.176 + __ movptr(obj, NULL_WORD); 1.177 + } else { 1.178 + __ movl(obj, val); 1.179 + // flatten object address if needed 1.180 + if (!precise || (obj.index() == noreg && obj.disp() == 0)) { 1.181 + __ store_check(obj.base()); 1.182 + } else { 1.183 + __ leal(rdx, obj); 1.184 + __ store_check(rdx); 1.185 + } 1.186 + } 1.187 + } 1.188 + break; 1.189 + case BarrierSet::ModRef: 1.190 + case BarrierSet::Other: 1.191 + if (val == noreg) { 1.192 + __ movptr(obj, NULL_WORD); 1.193 + } else { 1.194 + __ movl(obj, val); 1.195 + } 1.196 + break; 1.197 + default : 1.198 + ShouldNotReachHere(); 1.199 + 1.200 + } 1.201 +} 1.202 + 1.203 +Address TemplateTable::at_bcp(int offset) { 1.204 + assert(_desc->uses_bcp(), "inconsistent uses_bcp information"); 1.205 + return Address(rsi, offset); 1.206 +} 1.207 + 1.208 + 1.209 +void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg, 1.210 + Register temp_reg, bool load_bc_into_bc_reg/*=true*/, 1.211 + int byte_no) { 1.212 + if (!RewriteBytecodes) return; 1.213 + Label L_patch_done; 1.214 + 1.215 + switch (bc) { 1.216 + case Bytecodes::_fast_aputfield: 1.217 + case Bytecodes::_fast_bputfield: 1.218 + case Bytecodes::_fast_cputfield: 1.219 + case Bytecodes::_fast_dputfield: 1.220 + case Bytecodes::_fast_fputfield: 1.221 + case Bytecodes::_fast_iputfield: 1.222 + case Bytecodes::_fast_lputfield: 1.223 + case Bytecodes::_fast_sputfield: 1.224 + { 1.225 + // We skip bytecode quickening for putfield instructions when 1.226 + // the put_code written to the constant pool cache is zero. 1.227 + // This is required so that every execution of this instruction 1.228 + // calls out to InterpreterRuntime::resolve_get_put to do 1.229 + // additional, required work. 1.230 + assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range"); 1.231 + assert(load_bc_into_bc_reg, "we use bc_reg as temp"); 1.232 + __ get_cache_and_index_and_bytecode_at_bcp(bc_reg, temp_reg, temp_reg, byte_no, 1); 1.233 + __ movl(bc_reg, bc); 1.234 + __ cmpl(temp_reg, (int) 0); 1.235 + __ jcc(Assembler::zero, L_patch_done); // don't patch 1.236 + } 1.237 + break; 1.238 + default: 1.239 + assert(byte_no == -1, "sanity"); 1.240 + // the pair bytecodes have already done the load. 1.241 + if (load_bc_into_bc_reg) { 1.242 + __ movl(bc_reg, bc); 1.243 + } 1.244 + } 1.245 + 1.246 + if (JvmtiExport::can_post_breakpoint()) { 1.247 + Label L_fast_patch; 1.248 + // if a breakpoint is present we can't rewrite the stream directly 1.249 + __ movzbl(temp_reg, at_bcp(0)); 1.250 + __ cmpl(temp_reg, Bytecodes::_breakpoint); 1.251 + __ jcc(Assembler::notEqual, L_fast_patch); 1.252 + __ get_method(temp_reg); 1.253 + // Let breakpoint table handling rewrite to quicker bytecode 1.254 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rsi, bc_reg); 1.255 +#ifndef ASSERT 1.256 + __ jmpb(L_patch_done); 1.257 +#else 1.258 + __ jmp(L_patch_done); 1.259 +#endif 1.260 + __ bind(L_fast_patch); 1.261 + } 1.262 + 1.263 +#ifdef ASSERT 1.264 + Label L_okay; 1.265 + __ load_unsigned_byte(temp_reg, at_bcp(0)); 1.266 + __ cmpl(temp_reg, (int)Bytecodes::java_code(bc)); 1.267 + __ jccb(Assembler::equal, L_okay); 1.268 + __ cmpl(temp_reg, bc_reg); 1.269 + __ jcc(Assembler::equal, L_okay); 1.270 + __ stop("patching the wrong bytecode"); 1.271 + __ bind(L_okay); 1.272 +#endif 1.273 + 1.274 + // patch bytecode 1.275 + __ movb(at_bcp(0), bc_reg); 1.276 + __ bind(L_patch_done); 1.277 +} 1.278 + 1.279 +//---------------------------------------------------------------------------------------------------- 1.280 +// Individual instructions 1.281 + 1.282 +void TemplateTable::nop() { 1.283 + transition(vtos, vtos); 1.284 + // nothing to do 1.285 +} 1.286 + 1.287 +void TemplateTable::shouldnotreachhere() { 1.288 + transition(vtos, vtos); 1.289 + __ stop("shouldnotreachhere bytecode"); 1.290 +} 1.291 + 1.292 + 1.293 + 1.294 +void TemplateTable::aconst_null() { 1.295 + transition(vtos, atos); 1.296 + __ xorptr(rax, rax); 1.297 +} 1.298 + 1.299 + 1.300 +void TemplateTable::iconst(int value) { 1.301 + transition(vtos, itos); 1.302 + if (value == 0) { 1.303 + __ xorptr(rax, rax); 1.304 + } else { 1.305 + __ movptr(rax, value); 1.306 + } 1.307 +} 1.308 + 1.309 + 1.310 +void TemplateTable::lconst(int value) { 1.311 + transition(vtos, ltos); 1.312 + if (value == 0) { 1.313 + __ xorptr(rax, rax); 1.314 + } else { 1.315 + __ movptr(rax, value); 1.316 + } 1.317 + assert(value >= 0, "check this code"); 1.318 + __ xorptr(rdx, rdx); 1.319 +} 1.320 + 1.321 + 1.322 +void TemplateTable::fconst(int value) { 1.323 + transition(vtos, ftos); 1.324 + if (value == 0) { __ fldz(); 1.325 + } else if (value == 1) { __ fld1(); 1.326 + } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here 1.327 + } else { ShouldNotReachHere(); 1.328 + } 1.329 +} 1.330 + 1.331 + 1.332 +void TemplateTable::dconst(int value) { 1.333 + transition(vtos, dtos); 1.334 + if (value == 0) { __ fldz(); 1.335 + } else if (value == 1) { __ fld1(); 1.336 + } else { ShouldNotReachHere(); 1.337 + } 1.338 +} 1.339 + 1.340 + 1.341 +void TemplateTable::bipush() { 1.342 + transition(vtos, itos); 1.343 + __ load_signed_byte(rax, at_bcp(1)); 1.344 +} 1.345 + 1.346 + 1.347 +void TemplateTable::sipush() { 1.348 + transition(vtos, itos); 1.349 + __ load_unsigned_short(rax, at_bcp(1)); 1.350 + __ bswapl(rax); 1.351 + __ sarl(rax, 16); 1.352 +} 1.353 + 1.354 +void TemplateTable::ldc(bool wide) { 1.355 + transition(vtos, vtos); 1.356 + Label call_ldc, notFloat, notClass, Done; 1.357 + 1.358 + if (wide) { 1.359 + __ get_unsigned_2_byte_index_at_bcp(rbx, 1); 1.360 + } else { 1.361 + __ load_unsigned_byte(rbx, at_bcp(1)); 1.362 + } 1.363 + __ get_cpool_and_tags(rcx, rax); 1.364 + const int base_offset = ConstantPool::header_size() * wordSize; 1.365 + const int tags_offset = Array<u1>::base_offset_in_bytes(); 1.366 + 1.367 + // get type 1.368 + __ xorptr(rdx, rdx); 1.369 + __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset)); 1.370 + 1.371 + // unresolved class - get the resolved class 1.372 + __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass); 1.373 + __ jccb(Assembler::equal, call_ldc); 1.374 + 1.375 + // unresolved class in error (resolution failed) - call into runtime 1.376 + // so that the same error from first resolution attempt is thrown. 1.377 + __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError); 1.378 + __ jccb(Assembler::equal, call_ldc); 1.379 + 1.380 + // resolved class - need to call vm to get java mirror of the class 1.381 + __ cmpl(rdx, JVM_CONSTANT_Class); 1.382 + __ jcc(Assembler::notEqual, notClass); 1.383 + 1.384 + __ bind(call_ldc); 1.385 + __ movl(rcx, wide); 1.386 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx); 1.387 + __ push(atos); 1.388 + __ jmp(Done); 1.389 + 1.390 + __ bind(notClass); 1.391 + __ cmpl(rdx, JVM_CONSTANT_Float); 1.392 + __ jccb(Assembler::notEqual, notFloat); 1.393 + // ftos 1.394 + __ fld_s( Address(rcx, rbx, Address::times_ptr, base_offset)); 1.395 + __ push(ftos); 1.396 + __ jmp(Done); 1.397 + 1.398 + __ bind(notFloat); 1.399 +#ifdef ASSERT 1.400 + { Label L; 1.401 + __ cmpl(rdx, JVM_CONSTANT_Integer); 1.402 + __ jcc(Assembler::equal, L); 1.403 + // String and Object are rewritten to fast_aldc 1.404 + __ stop("unexpected tag type in ldc"); 1.405 + __ bind(L); 1.406 + } 1.407 +#endif 1.408 + // itos JVM_CONSTANT_Integer only 1.409 + __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset)); 1.410 + __ push(itos); 1.411 + __ bind(Done); 1.412 +} 1.413 + 1.414 +// Fast path for caching oop constants. 1.415 +void TemplateTable::fast_aldc(bool wide) { 1.416 + transition(vtos, atos); 1.417 + 1.418 + Register result = rax; 1.419 + Register tmp = rdx; 1.420 + int index_size = wide ? sizeof(u2) : sizeof(u1); 1.421 + 1.422 + Label resolved; 1.423 + 1.424 + // We are resolved if the resolved reference cache entry contains a 1.425 + // non-null object (String, MethodType, etc.) 1.426 + assert_different_registers(result, tmp); 1.427 + __ get_cache_index_at_bcp(tmp, 1, index_size); 1.428 + __ load_resolved_reference_at_index(result, tmp); 1.429 + __ testl(result, result); 1.430 + __ jcc(Assembler::notZero, resolved); 1.431 + 1.432 + address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); 1.433 + 1.434 + // first time invocation - must resolve first 1.435 + __ movl(tmp, (int)bytecode()); 1.436 + __ call_VM(result, entry, tmp); 1.437 + 1.438 + __ bind(resolved); 1.439 + 1.440 + if (VerifyOops) { 1.441 + __ verify_oop(result); 1.442 + } 1.443 +} 1.444 + 1.445 +void TemplateTable::ldc2_w() { 1.446 + transition(vtos, vtos); 1.447 + Label Long, Done; 1.448 + __ get_unsigned_2_byte_index_at_bcp(rbx, 1); 1.449 + 1.450 + __ get_cpool_and_tags(rcx, rax); 1.451 + const int base_offset = ConstantPool::header_size() * wordSize; 1.452 + const int tags_offset = Array<u1>::base_offset_in_bytes(); 1.453 + 1.454 + // get type 1.455 + __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double); 1.456 + __ jccb(Assembler::notEqual, Long); 1.457 + // dtos 1.458 + __ fld_d( Address(rcx, rbx, Address::times_ptr, base_offset)); 1.459 + __ push(dtos); 1.460 + __ jmpb(Done); 1.461 + 1.462 + __ bind(Long); 1.463 + // ltos 1.464 + __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize)); 1.465 + NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize))); 1.466 + 1.467 + __ push(ltos); 1.468 + 1.469 + __ bind(Done); 1.470 +} 1.471 + 1.472 + 1.473 +void TemplateTable::locals_index(Register reg, int offset) { 1.474 + __ load_unsigned_byte(reg, at_bcp(offset)); 1.475 + __ negptr(reg); 1.476 +} 1.477 + 1.478 + 1.479 +void TemplateTable::iload() { 1.480 + transition(vtos, itos); 1.481 + if (RewriteFrequentPairs) { 1.482 + Label rewrite, done; 1.483 + 1.484 + // get next byte 1.485 + __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload))); 1.486 + // if _iload, wait to rewrite to iload2. We only want to rewrite the 1.487 + // last two iloads in a pair. Comparing against fast_iload means that 1.488 + // the next bytecode is neither an iload or a caload, and therefore 1.489 + // an iload pair. 1.490 + __ cmpl(rbx, Bytecodes::_iload); 1.491 + __ jcc(Assembler::equal, done); 1.492 + 1.493 + __ cmpl(rbx, Bytecodes::_fast_iload); 1.494 + __ movl(rcx, Bytecodes::_fast_iload2); 1.495 + __ jccb(Assembler::equal, rewrite); 1.496 + 1.497 + // if _caload, rewrite to fast_icaload 1.498 + __ cmpl(rbx, Bytecodes::_caload); 1.499 + __ movl(rcx, Bytecodes::_fast_icaload); 1.500 + __ jccb(Assembler::equal, rewrite); 1.501 + 1.502 + // rewrite so iload doesn't check again. 1.503 + __ movl(rcx, Bytecodes::_fast_iload); 1.504 + 1.505 + // rewrite 1.506 + // rcx: fast bytecode 1.507 + __ bind(rewrite); 1.508 + patch_bytecode(Bytecodes::_iload, rcx, rbx, false); 1.509 + __ bind(done); 1.510 + } 1.511 + 1.512 + // Get the local value into tos 1.513 + locals_index(rbx); 1.514 + __ movl(rax, iaddress(rbx)); 1.515 +} 1.516 + 1.517 + 1.518 +void TemplateTable::fast_iload2() { 1.519 + transition(vtos, itos); 1.520 + locals_index(rbx); 1.521 + __ movl(rax, iaddress(rbx)); 1.522 + __ push(itos); 1.523 + locals_index(rbx, 3); 1.524 + __ movl(rax, iaddress(rbx)); 1.525 +} 1.526 + 1.527 +void TemplateTable::fast_iload() { 1.528 + transition(vtos, itos); 1.529 + locals_index(rbx); 1.530 + __ movl(rax, iaddress(rbx)); 1.531 +} 1.532 + 1.533 + 1.534 +void TemplateTable::lload() { 1.535 + transition(vtos, ltos); 1.536 + locals_index(rbx); 1.537 + __ movptr(rax, laddress(rbx)); 1.538 + NOT_LP64(__ movl(rdx, haddress(rbx))); 1.539 +} 1.540 + 1.541 + 1.542 +void TemplateTable::fload() { 1.543 + transition(vtos, ftos); 1.544 + locals_index(rbx); 1.545 + __ fld_s(faddress(rbx)); 1.546 +} 1.547 + 1.548 + 1.549 +void TemplateTable::dload() { 1.550 + transition(vtos, dtos); 1.551 + locals_index(rbx); 1.552 + __ fld_d(daddress(rbx)); 1.553 +} 1.554 + 1.555 + 1.556 +void TemplateTable::aload() { 1.557 + transition(vtos, atos); 1.558 + locals_index(rbx); 1.559 + __ movptr(rax, aaddress(rbx)); 1.560 +} 1.561 + 1.562 + 1.563 +void TemplateTable::locals_index_wide(Register reg) { 1.564 + __ load_unsigned_short(reg, at_bcp(2)); 1.565 + __ bswapl(reg); 1.566 + __ shrl(reg, 16); 1.567 + __ negptr(reg); 1.568 +} 1.569 + 1.570 + 1.571 +void TemplateTable::wide_iload() { 1.572 + transition(vtos, itos); 1.573 + locals_index_wide(rbx); 1.574 + __ movl(rax, iaddress(rbx)); 1.575 +} 1.576 + 1.577 + 1.578 +void TemplateTable::wide_lload() { 1.579 + transition(vtos, ltos); 1.580 + locals_index_wide(rbx); 1.581 + __ movptr(rax, laddress(rbx)); 1.582 + NOT_LP64(__ movl(rdx, haddress(rbx))); 1.583 +} 1.584 + 1.585 + 1.586 +void TemplateTable::wide_fload() { 1.587 + transition(vtos, ftos); 1.588 + locals_index_wide(rbx); 1.589 + __ fld_s(faddress(rbx)); 1.590 +} 1.591 + 1.592 + 1.593 +void TemplateTable::wide_dload() { 1.594 + transition(vtos, dtos); 1.595 + locals_index_wide(rbx); 1.596 + __ fld_d(daddress(rbx)); 1.597 +} 1.598 + 1.599 + 1.600 +void TemplateTable::wide_aload() { 1.601 + transition(vtos, atos); 1.602 + locals_index_wide(rbx); 1.603 + __ movptr(rax, aaddress(rbx)); 1.604 +} 1.605 + 1.606 +void TemplateTable::index_check(Register array, Register index) { 1.607 + // Pop ptr into array 1.608 + __ pop_ptr(array); 1.609 + index_check_without_pop(array, index); 1.610 +} 1.611 + 1.612 +void TemplateTable::index_check_without_pop(Register array, Register index) { 1.613 + // destroys rbx, 1.614 + // check array 1.615 + __ null_check(array, arrayOopDesc::length_offset_in_bytes()); 1.616 + LP64_ONLY(__ movslq(index, index)); 1.617 + // check index 1.618 + __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes())); 1.619 + if (index != rbx) { 1.620 + // ??? convention: move aberrant index into rbx, for exception message 1.621 + assert(rbx != array, "different registers"); 1.622 + __ mov(rbx, index); 1.623 + } 1.624 + __ jump_cc(Assembler::aboveEqual, 1.625 + ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry)); 1.626 +} 1.627 + 1.628 + 1.629 +void TemplateTable::iaload() { 1.630 + transition(itos, itos); 1.631 + // rdx: array 1.632 + index_check(rdx, rax); // kills rbx, 1.633 + // rax,: index 1.634 + __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT))); 1.635 +} 1.636 + 1.637 + 1.638 +void TemplateTable::laload() { 1.639 + transition(itos, ltos); 1.640 + // rax,: index 1.641 + // rdx: array 1.642 + index_check(rdx, rax); 1.643 + __ mov(rbx, rax); 1.644 + // rbx,: index 1.645 + __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize)); 1.646 + NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize))); 1.647 +} 1.648 + 1.649 + 1.650 +void TemplateTable::faload() { 1.651 + transition(itos, ftos); 1.652 + // rdx: array 1.653 + index_check(rdx, rax); // kills rbx, 1.654 + // rax,: index 1.655 + __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT))); 1.656 +} 1.657 + 1.658 + 1.659 +void TemplateTable::daload() { 1.660 + transition(itos, dtos); 1.661 + // rdx: array 1.662 + index_check(rdx, rax); // kills rbx, 1.663 + // rax,: index 1.664 + __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))); 1.665 +} 1.666 + 1.667 + 1.668 +void TemplateTable::aaload() { 1.669 + transition(itos, atos); 1.670 + // rdx: array 1.671 + index_check(rdx, rax); // kills rbx, 1.672 + // rax,: index 1.673 + __ movptr(rax, Address(rdx, rax, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 1.674 +} 1.675 + 1.676 + 1.677 +void TemplateTable::baload() { 1.678 + transition(itos, itos); 1.679 + // rdx: array 1.680 + index_check(rdx, rax); // kills rbx, 1.681 + // rax,: index 1.682 + // can do better code for P5 - fix this at some point 1.683 + __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE))); 1.684 + __ mov(rax, rbx); 1.685 +} 1.686 + 1.687 + 1.688 +void TemplateTable::caload() { 1.689 + transition(itos, itos); 1.690 + // rdx: array 1.691 + index_check(rdx, rax); // kills rbx, 1.692 + // rax,: index 1.693 + // can do better code for P5 - may want to improve this at some point 1.694 + __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR))); 1.695 + __ mov(rax, rbx); 1.696 +} 1.697 + 1.698 +// iload followed by caload frequent pair 1.699 +void TemplateTable::fast_icaload() { 1.700 + transition(vtos, itos); 1.701 + // load index out of locals 1.702 + locals_index(rbx); 1.703 + __ movl(rax, iaddress(rbx)); 1.704 + 1.705 + // rdx: array 1.706 + index_check(rdx, rax); 1.707 + // rax,: index 1.708 + __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR))); 1.709 + __ mov(rax, rbx); 1.710 +} 1.711 + 1.712 +void TemplateTable::saload() { 1.713 + transition(itos, itos); 1.714 + // rdx: array 1.715 + index_check(rdx, rax); // kills rbx, 1.716 + // rax,: index 1.717 + // can do better code for P5 - may want to improve this at some point 1.718 + __ load_signed_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT))); 1.719 + __ mov(rax, rbx); 1.720 +} 1.721 + 1.722 + 1.723 +void TemplateTable::iload(int n) { 1.724 + transition(vtos, itos); 1.725 + __ movl(rax, iaddress(n)); 1.726 +} 1.727 + 1.728 + 1.729 +void TemplateTable::lload(int n) { 1.730 + transition(vtos, ltos); 1.731 + __ movptr(rax, laddress(n)); 1.732 + NOT_LP64(__ movptr(rdx, haddress(n))); 1.733 +} 1.734 + 1.735 + 1.736 +void TemplateTable::fload(int n) { 1.737 + transition(vtos, ftos); 1.738 + __ fld_s(faddress(n)); 1.739 +} 1.740 + 1.741 + 1.742 +void TemplateTable::dload(int n) { 1.743 + transition(vtos, dtos); 1.744 + __ fld_d(daddress(n)); 1.745 +} 1.746 + 1.747 + 1.748 +void TemplateTable::aload(int n) { 1.749 + transition(vtos, atos); 1.750 + __ movptr(rax, aaddress(n)); 1.751 +} 1.752 + 1.753 + 1.754 +void TemplateTable::aload_0() { 1.755 + transition(vtos, atos); 1.756 + // According to bytecode histograms, the pairs: 1.757 + // 1.758 + // _aload_0, _fast_igetfield 1.759 + // _aload_0, _fast_agetfield 1.760 + // _aload_0, _fast_fgetfield 1.761 + // 1.762 + // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0 1.763 + // bytecode checks if the next bytecode is either _fast_igetfield, 1.764 + // _fast_agetfield or _fast_fgetfield and then rewrites the 1.765 + // current bytecode into a pair bytecode; otherwise it rewrites the current 1.766 + // bytecode into _fast_aload_0 that doesn't do the pair check anymore. 1.767 + // 1.768 + // Note: If the next bytecode is _getfield, the rewrite must be delayed, 1.769 + // otherwise we may miss an opportunity for a pair. 1.770 + // 1.771 + // Also rewrite frequent pairs 1.772 + // aload_0, aload_1 1.773 + // aload_0, iload_1 1.774 + // These bytecodes with a small amount of code are most profitable to rewrite 1.775 + if (RewriteFrequentPairs) { 1.776 + Label rewrite, done; 1.777 + // get next byte 1.778 + __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0))); 1.779 + 1.780 + // do actual aload_0 1.781 + aload(0); 1.782 + 1.783 + // if _getfield then wait with rewrite 1.784 + __ cmpl(rbx, Bytecodes::_getfield); 1.785 + __ jcc(Assembler::equal, done); 1.786 + 1.787 + // if _igetfield then reqrite to _fast_iaccess_0 1.788 + assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 1.789 + __ cmpl(rbx, Bytecodes::_fast_igetfield); 1.790 + __ movl(rcx, Bytecodes::_fast_iaccess_0); 1.791 + __ jccb(Assembler::equal, rewrite); 1.792 + 1.793 + // if _agetfield then reqrite to _fast_aaccess_0 1.794 + assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 1.795 + __ cmpl(rbx, Bytecodes::_fast_agetfield); 1.796 + __ movl(rcx, Bytecodes::_fast_aaccess_0); 1.797 + __ jccb(Assembler::equal, rewrite); 1.798 + 1.799 + // if _fgetfield then reqrite to _fast_faccess_0 1.800 + assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition"); 1.801 + __ cmpl(rbx, Bytecodes::_fast_fgetfield); 1.802 + __ movl(rcx, Bytecodes::_fast_faccess_0); 1.803 + __ jccb(Assembler::equal, rewrite); 1.804 + 1.805 + // else rewrite to _fast_aload0 1.806 + assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition"); 1.807 + __ movl(rcx, Bytecodes::_fast_aload_0); 1.808 + 1.809 + // rewrite 1.810 + // rcx: fast bytecode 1.811 + __ bind(rewrite); 1.812 + patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false); 1.813 + 1.814 + __ bind(done); 1.815 + } else { 1.816 + aload(0); 1.817 + } 1.818 +} 1.819 + 1.820 +void TemplateTable::istore() { 1.821 + transition(itos, vtos); 1.822 + locals_index(rbx); 1.823 + __ movl(iaddress(rbx), rax); 1.824 +} 1.825 + 1.826 + 1.827 +void TemplateTable::lstore() { 1.828 + transition(ltos, vtos); 1.829 + locals_index(rbx); 1.830 + __ movptr(laddress(rbx), rax); 1.831 + NOT_LP64(__ movptr(haddress(rbx), rdx)); 1.832 +} 1.833 + 1.834 + 1.835 +void TemplateTable::fstore() { 1.836 + transition(ftos, vtos); 1.837 + locals_index(rbx); 1.838 + __ fstp_s(faddress(rbx)); 1.839 +} 1.840 + 1.841 + 1.842 +void TemplateTable::dstore() { 1.843 + transition(dtos, vtos); 1.844 + locals_index(rbx); 1.845 + __ fstp_d(daddress(rbx)); 1.846 +} 1.847 + 1.848 + 1.849 +void TemplateTable::astore() { 1.850 + transition(vtos, vtos); 1.851 + __ pop_ptr(rax); 1.852 + locals_index(rbx); 1.853 + __ movptr(aaddress(rbx), rax); 1.854 +} 1.855 + 1.856 + 1.857 +void TemplateTable::wide_istore() { 1.858 + transition(vtos, vtos); 1.859 + __ pop_i(rax); 1.860 + locals_index_wide(rbx); 1.861 + __ movl(iaddress(rbx), rax); 1.862 +} 1.863 + 1.864 + 1.865 +void TemplateTable::wide_lstore() { 1.866 + transition(vtos, vtos); 1.867 + __ pop_l(rax, rdx); 1.868 + locals_index_wide(rbx); 1.869 + __ movptr(laddress(rbx), rax); 1.870 + NOT_LP64(__ movl(haddress(rbx), rdx)); 1.871 +} 1.872 + 1.873 + 1.874 +void TemplateTable::wide_fstore() { 1.875 + wide_istore(); 1.876 +} 1.877 + 1.878 + 1.879 +void TemplateTable::wide_dstore() { 1.880 + wide_lstore(); 1.881 +} 1.882 + 1.883 + 1.884 +void TemplateTable::wide_astore() { 1.885 + transition(vtos, vtos); 1.886 + __ pop_ptr(rax); 1.887 + locals_index_wide(rbx); 1.888 + __ movptr(aaddress(rbx), rax); 1.889 +} 1.890 + 1.891 + 1.892 +void TemplateTable::iastore() { 1.893 + transition(itos, vtos); 1.894 + __ pop_i(rbx); 1.895 + // rax,: value 1.896 + // rdx: array 1.897 + index_check(rdx, rbx); // prefer index in rbx, 1.898 + // rbx,: index 1.899 + __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax); 1.900 +} 1.901 + 1.902 + 1.903 +void TemplateTable::lastore() { 1.904 + transition(ltos, vtos); 1.905 + __ pop_i(rbx); 1.906 + // rax,: low(value) 1.907 + // rcx: array 1.908 + // rdx: high(value) 1.909 + index_check(rcx, rbx); // prefer index in rbx, 1.910 + // rbx,: index 1.911 + __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax); 1.912 + NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx)); 1.913 +} 1.914 + 1.915 + 1.916 +void TemplateTable::fastore() { 1.917 + transition(ftos, vtos); 1.918 + __ pop_i(rbx); 1.919 + // rdx: array 1.920 + // st0: value 1.921 + index_check(rdx, rbx); // prefer index in rbx, 1.922 + // rbx,: index 1.923 + __ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT))); 1.924 +} 1.925 + 1.926 + 1.927 +void TemplateTable::dastore() { 1.928 + transition(dtos, vtos); 1.929 + __ pop_i(rbx); 1.930 + // rdx: array 1.931 + // st0: value 1.932 + index_check(rdx, rbx); // prefer index in rbx, 1.933 + // rbx,: index 1.934 + __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))); 1.935 +} 1.936 + 1.937 + 1.938 +void TemplateTable::aastore() { 1.939 + Label is_null, ok_is_subtype, done; 1.940 + transition(vtos, vtos); 1.941 + // stack: ..., array, index, value 1.942 + __ movptr(rax, at_tos()); // Value 1.943 + __ movl(rcx, at_tos_p1()); // Index 1.944 + __ movptr(rdx, at_tos_p2()); // Array 1.945 + 1.946 + Address element_address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 1.947 + index_check_without_pop(rdx, rcx); // kills rbx, 1.948 + // do array store check - check for NULL value first 1.949 + __ testptr(rax, rax); 1.950 + __ jcc(Assembler::zero, is_null); 1.951 + 1.952 + // Move subklass into EBX 1.953 + __ load_klass(rbx, rax); 1.954 + // Move superklass into EAX 1.955 + __ load_klass(rax, rdx); 1.956 + __ movptr(rax, Address(rax, ObjArrayKlass::element_klass_offset())); 1.957 + // Compress array+index*wordSize+12 into a single register. Frees ECX. 1.958 + __ lea(rdx, element_address); 1.959 + 1.960 + // Generate subtype check. Blows ECX. Resets EDI to locals. 1.961 + // Superklass in EAX. Subklass in EBX. 1.962 + __ gen_subtype_check( rbx, ok_is_subtype ); 1.963 + 1.964 + // Come here on failure 1.965 + // object is at TOS 1.966 + __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry)); 1.967 + 1.968 + // Come here on success 1.969 + __ bind(ok_is_subtype); 1.970 + 1.971 + // Get the value to store 1.972 + __ movptr(rax, at_rsp()); 1.973 + // and store it with appropriate barrier 1.974 + do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true); 1.975 + 1.976 + __ jmp(done); 1.977 + 1.978 + // Have a NULL in EAX, EDX=array, ECX=index. Store NULL at ary[idx] 1.979 + __ bind(is_null); 1.980 + __ profile_null_seen(rbx); 1.981 + 1.982 + // Store NULL, (noreg means NULL to do_oop_store) 1.983 + do_oop_store(_masm, element_address, noreg, _bs->kind(), true); 1.984 + 1.985 + // Pop stack arguments 1.986 + __ bind(done); 1.987 + __ addptr(rsp, 3 * Interpreter::stackElementSize); 1.988 +} 1.989 + 1.990 + 1.991 +void TemplateTable::bastore() { 1.992 + transition(itos, vtos); 1.993 + __ pop_i(rbx); 1.994 + // rax,: value 1.995 + // rdx: array 1.996 + index_check(rdx, rbx); // prefer index in rbx, 1.997 + // rbx,: index 1.998 + __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax); 1.999 +} 1.1000 + 1.1001 + 1.1002 +void TemplateTable::castore() { 1.1003 + transition(itos, vtos); 1.1004 + __ pop_i(rbx); 1.1005 + // rax,: value 1.1006 + // rdx: array 1.1007 + index_check(rdx, rbx); // prefer index in rbx, 1.1008 + // rbx,: index 1.1009 + __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax); 1.1010 +} 1.1011 + 1.1012 + 1.1013 +void TemplateTable::sastore() { 1.1014 + castore(); 1.1015 +} 1.1016 + 1.1017 + 1.1018 +void TemplateTable::istore(int n) { 1.1019 + transition(itos, vtos); 1.1020 + __ movl(iaddress(n), rax); 1.1021 +} 1.1022 + 1.1023 + 1.1024 +void TemplateTable::lstore(int n) { 1.1025 + transition(ltos, vtos); 1.1026 + __ movptr(laddress(n), rax); 1.1027 + NOT_LP64(__ movptr(haddress(n), rdx)); 1.1028 +} 1.1029 + 1.1030 + 1.1031 +void TemplateTable::fstore(int n) { 1.1032 + transition(ftos, vtos); 1.1033 + __ fstp_s(faddress(n)); 1.1034 +} 1.1035 + 1.1036 + 1.1037 +void TemplateTable::dstore(int n) { 1.1038 + transition(dtos, vtos); 1.1039 + __ fstp_d(daddress(n)); 1.1040 +} 1.1041 + 1.1042 + 1.1043 +void TemplateTable::astore(int n) { 1.1044 + transition(vtos, vtos); 1.1045 + __ pop_ptr(rax); 1.1046 + __ movptr(aaddress(n), rax); 1.1047 +} 1.1048 + 1.1049 + 1.1050 +void TemplateTable::pop() { 1.1051 + transition(vtos, vtos); 1.1052 + __ addptr(rsp, Interpreter::stackElementSize); 1.1053 +} 1.1054 + 1.1055 + 1.1056 +void TemplateTable::pop2() { 1.1057 + transition(vtos, vtos); 1.1058 + __ addptr(rsp, 2*Interpreter::stackElementSize); 1.1059 +} 1.1060 + 1.1061 + 1.1062 +void TemplateTable::dup() { 1.1063 + transition(vtos, vtos); 1.1064 + // stack: ..., a 1.1065 + __ load_ptr(0, rax); 1.1066 + __ push_ptr(rax); 1.1067 + // stack: ..., a, a 1.1068 +} 1.1069 + 1.1070 + 1.1071 +void TemplateTable::dup_x1() { 1.1072 + transition(vtos, vtos); 1.1073 + // stack: ..., a, b 1.1074 + __ load_ptr( 0, rax); // load b 1.1075 + __ load_ptr( 1, rcx); // load a 1.1076 + __ store_ptr(1, rax); // store b 1.1077 + __ store_ptr(0, rcx); // store a 1.1078 + __ push_ptr(rax); // push b 1.1079 + // stack: ..., b, a, b 1.1080 +} 1.1081 + 1.1082 + 1.1083 +void TemplateTable::dup_x2() { 1.1084 + transition(vtos, vtos); 1.1085 + // stack: ..., a, b, c 1.1086 + __ load_ptr( 0, rax); // load c 1.1087 + __ load_ptr( 2, rcx); // load a 1.1088 + __ store_ptr(2, rax); // store c in a 1.1089 + __ push_ptr(rax); // push c 1.1090 + // stack: ..., c, b, c, c 1.1091 + __ load_ptr( 2, rax); // load b 1.1092 + __ store_ptr(2, rcx); // store a in b 1.1093 + // stack: ..., c, a, c, c 1.1094 + __ store_ptr(1, rax); // store b in c 1.1095 + // stack: ..., c, a, b, c 1.1096 +} 1.1097 + 1.1098 + 1.1099 +void TemplateTable::dup2() { 1.1100 + transition(vtos, vtos); 1.1101 + // stack: ..., a, b 1.1102 + __ load_ptr(1, rax); // load a 1.1103 + __ push_ptr(rax); // push a 1.1104 + __ load_ptr(1, rax); // load b 1.1105 + __ push_ptr(rax); // push b 1.1106 + // stack: ..., a, b, a, b 1.1107 +} 1.1108 + 1.1109 + 1.1110 +void TemplateTable::dup2_x1() { 1.1111 + transition(vtos, vtos); 1.1112 + // stack: ..., a, b, c 1.1113 + __ load_ptr( 0, rcx); // load c 1.1114 + __ load_ptr( 1, rax); // load b 1.1115 + __ push_ptr(rax); // push b 1.1116 + __ push_ptr(rcx); // push c 1.1117 + // stack: ..., a, b, c, b, c 1.1118 + __ store_ptr(3, rcx); // store c in b 1.1119 + // stack: ..., a, c, c, b, c 1.1120 + __ load_ptr( 4, rcx); // load a 1.1121 + __ store_ptr(2, rcx); // store a in 2nd c 1.1122 + // stack: ..., a, c, a, b, c 1.1123 + __ store_ptr(4, rax); // store b in a 1.1124 + // stack: ..., b, c, a, b, c 1.1125 + // stack: ..., b, c, a, b, c 1.1126 +} 1.1127 + 1.1128 + 1.1129 +void TemplateTable::dup2_x2() { 1.1130 + transition(vtos, vtos); 1.1131 + // stack: ..., a, b, c, d 1.1132 + __ load_ptr( 0, rcx); // load d 1.1133 + __ load_ptr( 1, rax); // load c 1.1134 + __ push_ptr(rax); // push c 1.1135 + __ push_ptr(rcx); // push d 1.1136 + // stack: ..., a, b, c, d, c, d 1.1137 + __ load_ptr( 4, rax); // load b 1.1138 + __ store_ptr(2, rax); // store b in d 1.1139 + __ store_ptr(4, rcx); // store d in b 1.1140 + // stack: ..., a, d, c, b, c, d 1.1141 + __ load_ptr( 5, rcx); // load a 1.1142 + __ load_ptr( 3, rax); // load c 1.1143 + __ store_ptr(3, rcx); // store a in c 1.1144 + __ store_ptr(5, rax); // store c in a 1.1145 + // stack: ..., c, d, a, b, c, d 1.1146 + // stack: ..., c, d, a, b, c, d 1.1147 +} 1.1148 + 1.1149 + 1.1150 +void TemplateTable::swap() { 1.1151 + transition(vtos, vtos); 1.1152 + // stack: ..., a, b 1.1153 + __ load_ptr( 1, rcx); // load a 1.1154 + __ load_ptr( 0, rax); // load b 1.1155 + __ store_ptr(0, rcx); // store a in b 1.1156 + __ store_ptr(1, rax); // store b in a 1.1157 + // stack: ..., b, a 1.1158 +} 1.1159 + 1.1160 + 1.1161 +void TemplateTable::iop2(Operation op) { 1.1162 + transition(itos, itos); 1.1163 + switch (op) { 1.1164 + case add : __ pop_i(rdx); __ addl (rax, rdx); break; 1.1165 + case sub : __ mov(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break; 1.1166 + case mul : __ pop_i(rdx); __ imull(rax, rdx); break; 1.1167 + case _and : __ pop_i(rdx); __ andl (rax, rdx); break; 1.1168 + case _or : __ pop_i(rdx); __ orl (rax, rdx); break; 1.1169 + case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break; 1.1170 + case shl : __ mov(rcx, rax); __ pop_i(rax); __ shll (rax); break; // implicit masking of lower 5 bits by Intel shift instr. 1.1171 + case shr : __ mov(rcx, rax); __ pop_i(rax); __ sarl (rax); break; // implicit masking of lower 5 bits by Intel shift instr. 1.1172 + case ushr : __ mov(rcx, rax); __ pop_i(rax); __ shrl (rax); break; // implicit masking of lower 5 bits by Intel shift instr. 1.1173 + default : ShouldNotReachHere(); 1.1174 + } 1.1175 +} 1.1176 + 1.1177 + 1.1178 +void TemplateTable::lop2(Operation op) { 1.1179 + transition(ltos, ltos); 1.1180 + __ pop_l(rbx, rcx); 1.1181 + switch (op) { 1.1182 + case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break; 1.1183 + case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx); 1.1184 + __ mov (rax, rbx); __ mov (rdx, rcx); break; 1.1185 + case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break; 1.1186 + case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break; 1.1187 + case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break; 1.1188 + default : ShouldNotReachHere(); 1.1189 + } 1.1190 +} 1.1191 + 1.1192 + 1.1193 +void TemplateTable::idiv() { 1.1194 + transition(itos, itos); 1.1195 + __ mov(rcx, rax); 1.1196 + __ pop_i(rax); 1.1197 + // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If 1.1198 + // they are not equal, one could do a normal division (no correction 1.1199 + // needed), which may speed up this implementation for the common case. 1.1200 + // (see also JVM spec., p.243 & p.271) 1.1201 + __ corrected_idivl(rcx); 1.1202 +} 1.1203 + 1.1204 + 1.1205 +void TemplateTable::irem() { 1.1206 + transition(itos, itos); 1.1207 + __ mov(rcx, rax); 1.1208 + __ pop_i(rax); 1.1209 + // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If 1.1210 + // they are not equal, one could do a normal division (no correction 1.1211 + // needed), which may speed up this implementation for the common case. 1.1212 + // (see also JVM spec., p.243 & p.271) 1.1213 + __ corrected_idivl(rcx); 1.1214 + __ mov(rax, rdx); 1.1215 +} 1.1216 + 1.1217 + 1.1218 +void TemplateTable::lmul() { 1.1219 + transition(ltos, ltos); 1.1220 + __ pop_l(rbx, rcx); 1.1221 + __ push(rcx); __ push(rbx); 1.1222 + __ push(rdx); __ push(rax); 1.1223 + __ lmul(2 * wordSize, 0); 1.1224 + __ addptr(rsp, 4 * wordSize); // take off temporaries 1.1225 +} 1.1226 + 1.1227 + 1.1228 +void TemplateTable::ldiv() { 1.1229 + transition(ltos, ltos); 1.1230 + __ pop_l(rbx, rcx); 1.1231 + __ push(rcx); __ push(rbx); 1.1232 + __ push(rdx); __ push(rax); 1.1233 + // check if y = 0 1.1234 + __ orl(rax, rdx); 1.1235 + __ jump_cc(Assembler::zero, 1.1236 + ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1.1237 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv)); 1.1238 + __ addptr(rsp, 4 * wordSize); // take off temporaries 1.1239 +} 1.1240 + 1.1241 + 1.1242 +void TemplateTable::lrem() { 1.1243 + transition(ltos, ltos); 1.1244 + __ pop_l(rbx, rcx); 1.1245 + __ push(rcx); __ push(rbx); 1.1246 + __ push(rdx); __ push(rax); 1.1247 + // check if y = 0 1.1248 + __ orl(rax, rdx); 1.1249 + __ jump_cc(Assembler::zero, 1.1250 + ExternalAddress(Interpreter::_throw_ArithmeticException_entry)); 1.1251 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem)); 1.1252 + __ addptr(rsp, 4 * wordSize); 1.1253 +} 1.1254 + 1.1255 + 1.1256 +void TemplateTable::lshl() { 1.1257 + transition(itos, ltos); 1.1258 + __ movl(rcx, rax); // get shift count 1.1259 + __ pop_l(rax, rdx); // get shift value 1.1260 + __ lshl(rdx, rax); 1.1261 +} 1.1262 + 1.1263 + 1.1264 +void TemplateTable::lshr() { 1.1265 + transition(itos, ltos); 1.1266 + __ mov(rcx, rax); // get shift count 1.1267 + __ pop_l(rax, rdx); // get shift value 1.1268 + __ lshr(rdx, rax, true); 1.1269 +} 1.1270 + 1.1271 + 1.1272 +void TemplateTable::lushr() { 1.1273 + transition(itos, ltos); 1.1274 + __ mov(rcx, rax); // get shift count 1.1275 + __ pop_l(rax, rdx); // get shift value 1.1276 + __ lshr(rdx, rax); 1.1277 +} 1.1278 + 1.1279 + 1.1280 +void TemplateTable::fop2(Operation op) { 1.1281 + transition(ftos, ftos); 1.1282 + switch (op) { 1.1283 + case add: __ fadd_s (at_rsp()); break; 1.1284 + case sub: __ fsubr_s(at_rsp()); break; 1.1285 + case mul: __ fmul_s (at_rsp()); break; 1.1286 + case div: __ fdivr_s(at_rsp()); break; 1.1287 + case rem: __ fld_s (at_rsp()); __ fremr(rax); break; 1.1288 + default : ShouldNotReachHere(); 1.1289 + } 1.1290 + __ f2ieee(); 1.1291 + __ pop(rax); // pop float thing off 1.1292 +} 1.1293 + 1.1294 + 1.1295 +void TemplateTable::dop2(Operation op) { 1.1296 + transition(dtos, dtos); 1.1297 + 1.1298 + switch (op) { 1.1299 + case add: __ fadd_d (at_rsp()); break; 1.1300 + case sub: __ fsubr_d(at_rsp()); break; 1.1301 + case mul: { 1.1302 + Label L_strict; 1.1303 + Label L_join; 1.1304 + const Address access_flags (rcx, Method::access_flags_offset()); 1.1305 + __ get_method(rcx); 1.1306 + __ movl(rcx, access_flags); 1.1307 + __ testl(rcx, JVM_ACC_STRICT); 1.1308 + __ jccb(Assembler::notZero, L_strict); 1.1309 + __ fmul_d (at_rsp()); 1.1310 + __ jmpb(L_join); 1.1311 + __ bind(L_strict); 1.1312 + __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1())); 1.1313 + __ fmulp(); 1.1314 + __ fmul_d (at_rsp()); 1.1315 + __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2())); 1.1316 + __ fmulp(); 1.1317 + __ bind(L_join); 1.1318 + break; 1.1319 + } 1.1320 + case div: { 1.1321 + Label L_strict; 1.1322 + Label L_join; 1.1323 + const Address access_flags (rcx, Method::access_flags_offset()); 1.1324 + __ get_method(rcx); 1.1325 + __ movl(rcx, access_flags); 1.1326 + __ testl(rcx, JVM_ACC_STRICT); 1.1327 + __ jccb(Assembler::notZero, L_strict); 1.1328 + __ fdivr_d(at_rsp()); 1.1329 + __ jmp(L_join); 1.1330 + __ bind(L_strict); 1.1331 + __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1())); 1.1332 + __ fmul_d (at_rsp()); 1.1333 + __ fdivrp(); 1.1334 + __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2())); 1.1335 + __ fmulp(); 1.1336 + __ bind(L_join); 1.1337 + break; 1.1338 + } 1.1339 + case rem: __ fld_d (at_rsp()); __ fremr(rax); break; 1.1340 + default : ShouldNotReachHere(); 1.1341 + } 1.1342 + __ d2ieee(); 1.1343 + // Pop double precision number from rsp. 1.1344 + __ pop(rax); 1.1345 + __ pop(rdx); 1.1346 +} 1.1347 + 1.1348 + 1.1349 +void TemplateTable::ineg() { 1.1350 + transition(itos, itos); 1.1351 + __ negl(rax); 1.1352 +} 1.1353 + 1.1354 + 1.1355 +void TemplateTable::lneg() { 1.1356 + transition(ltos, ltos); 1.1357 + __ lneg(rdx, rax); 1.1358 +} 1.1359 + 1.1360 + 1.1361 +void TemplateTable::fneg() { 1.1362 + transition(ftos, ftos); 1.1363 + __ fchs(); 1.1364 +} 1.1365 + 1.1366 + 1.1367 +void TemplateTable::dneg() { 1.1368 + transition(dtos, dtos); 1.1369 + __ fchs(); 1.1370 +} 1.1371 + 1.1372 + 1.1373 +void TemplateTable::iinc() { 1.1374 + transition(vtos, vtos); 1.1375 + __ load_signed_byte(rdx, at_bcp(2)); // get constant 1.1376 + locals_index(rbx); 1.1377 + __ addl(iaddress(rbx), rdx); 1.1378 +} 1.1379 + 1.1380 + 1.1381 +void TemplateTable::wide_iinc() { 1.1382 + transition(vtos, vtos); 1.1383 + __ movl(rdx, at_bcp(4)); // get constant 1.1384 + locals_index_wide(rbx); 1.1385 + __ bswapl(rdx); // swap bytes & sign-extend constant 1.1386 + __ sarl(rdx, 16); 1.1387 + __ addl(iaddress(rbx), rdx); 1.1388 + // Note: should probably use only one movl to get both 1.1389 + // the index and the constant -> fix this 1.1390 +} 1.1391 + 1.1392 + 1.1393 +void TemplateTable::convert() { 1.1394 + // Checking 1.1395 +#ifdef ASSERT 1.1396 + { TosState tos_in = ilgl; 1.1397 + TosState tos_out = ilgl; 1.1398 + switch (bytecode()) { 1.1399 + case Bytecodes::_i2l: // fall through 1.1400 + case Bytecodes::_i2f: // fall through 1.1401 + case Bytecodes::_i2d: // fall through 1.1402 + case Bytecodes::_i2b: // fall through 1.1403 + case Bytecodes::_i2c: // fall through 1.1404 + case Bytecodes::_i2s: tos_in = itos; break; 1.1405 + case Bytecodes::_l2i: // fall through 1.1406 + case Bytecodes::_l2f: // fall through 1.1407 + case Bytecodes::_l2d: tos_in = ltos; break; 1.1408 + case Bytecodes::_f2i: // fall through 1.1409 + case Bytecodes::_f2l: // fall through 1.1410 + case Bytecodes::_f2d: tos_in = ftos; break; 1.1411 + case Bytecodes::_d2i: // fall through 1.1412 + case Bytecodes::_d2l: // fall through 1.1413 + case Bytecodes::_d2f: tos_in = dtos; break; 1.1414 + default : ShouldNotReachHere(); 1.1415 + } 1.1416 + switch (bytecode()) { 1.1417 + case Bytecodes::_l2i: // fall through 1.1418 + case Bytecodes::_f2i: // fall through 1.1419 + case Bytecodes::_d2i: // fall through 1.1420 + case Bytecodes::_i2b: // fall through 1.1421 + case Bytecodes::_i2c: // fall through 1.1422 + case Bytecodes::_i2s: tos_out = itos; break; 1.1423 + case Bytecodes::_i2l: // fall through 1.1424 + case Bytecodes::_f2l: // fall through 1.1425 + case Bytecodes::_d2l: tos_out = ltos; break; 1.1426 + case Bytecodes::_i2f: // fall through 1.1427 + case Bytecodes::_l2f: // fall through 1.1428 + case Bytecodes::_d2f: tos_out = ftos; break; 1.1429 + case Bytecodes::_i2d: // fall through 1.1430 + case Bytecodes::_l2d: // fall through 1.1431 + case Bytecodes::_f2d: tos_out = dtos; break; 1.1432 + default : ShouldNotReachHere(); 1.1433 + } 1.1434 + transition(tos_in, tos_out); 1.1435 + } 1.1436 +#endif // ASSERT 1.1437 + 1.1438 + // Conversion 1.1439 + // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation) 1.1440 + switch (bytecode()) { 1.1441 + case Bytecodes::_i2l: 1.1442 + __ extend_sign(rdx, rax); 1.1443 + break; 1.1444 + case Bytecodes::_i2f: 1.1445 + __ push(rax); // store int on tos 1.1446 + __ fild_s(at_rsp()); // load int to ST0 1.1447 + __ f2ieee(); // truncate to float size 1.1448 + __ pop(rcx); // adjust rsp 1.1449 + break; 1.1450 + case Bytecodes::_i2d: 1.1451 + __ push(rax); // add one slot for d2ieee() 1.1452 + __ push(rax); // store int on tos 1.1453 + __ fild_s(at_rsp()); // load int to ST0 1.1454 + __ d2ieee(); // truncate to double size 1.1455 + __ pop(rcx); // adjust rsp 1.1456 + __ pop(rcx); 1.1457 + break; 1.1458 + case Bytecodes::_i2b: 1.1459 + __ shll(rax, 24); // truncate upper 24 bits 1.1460 + __ sarl(rax, 24); // and sign-extend byte 1.1461 + LP64_ONLY(__ movsbl(rax, rax)); 1.1462 + break; 1.1463 + case Bytecodes::_i2c: 1.1464 + __ andl(rax, 0xFFFF); // truncate upper 16 bits 1.1465 + LP64_ONLY(__ movzwl(rax, rax)); 1.1466 + break; 1.1467 + case Bytecodes::_i2s: 1.1468 + __ shll(rax, 16); // truncate upper 16 bits 1.1469 + __ sarl(rax, 16); // and sign-extend short 1.1470 + LP64_ONLY(__ movswl(rax, rax)); 1.1471 + break; 1.1472 + case Bytecodes::_l2i: 1.1473 + /* nothing to do */ 1.1474 + break; 1.1475 + case Bytecodes::_l2f: 1.1476 + __ push(rdx); // store long on tos 1.1477 + __ push(rax); 1.1478 + __ fild_d(at_rsp()); // load long to ST0 1.1479 + __ f2ieee(); // truncate to float size 1.1480 + __ pop(rcx); // adjust rsp 1.1481 + __ pop(rcx); 1.1482 + break; 1.1483 + case Bytecodes::_l2d: 1.1484 + __ push(rdx); // store long on tos 1.1485 + __ push(rax); 1.1486 + __ fild_d(at_rsp()); // load long to ST0 1.1487 + __ d2ieee(); // truncate to double size 1.1488 + __ pop(rcx); // adjust rsp 1.1489 + __ pop(rcx); 1.1490 + break; 1.1491 + case Bytecodes::_f2i: 1.1492 + __ push(rcx); // reserve space for argument 1.1493 + __ fstp_s(at_rsp()); // pass float argument on stack 1.1494 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1); 1.1495 + break; 1.1496 + case Bytecodes::_f2l: 1.1497 + __ push(rcx); // reserve space for argument 1.1498 + __ fstp_s(at_rsp()); // pass float argument on stack 1.1499 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1); 1.1500 + break; 1.1501 + case Bytecodes::_f2d: 1.1502 + /* nothing to do */ 1.1503 + break; 1.1504 + case Bytecodes::_d2i: 1.1505 + __ push(rcx); // reserve space for argument 1.1506 + __ push(rcx); 1.1507 + __ fstp_d(at_rsp()); // pass double argument on stack 1.1508 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2); 1.1509 + break; 1.1510 + case Bytecodes::_d2l: 1.1511 + __ push(rcx); // reserve space for argument 1.1512 + __ push(rcx); 1.1513 + __ fstp_d(at_rsp()); // pass double argument on stack 1.1514 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2); 1.1515 + break; 1.1516 + case Bytecodes::_d2f: 1.1517 + __ push(rcx); // reserve space for f2ieee() 1.1518 + __ f2ieee(); // truncate to float size 1.1519 + __ pop(rcx); // adjust rsp 1.1520 + break; 1.1521 + default : 1.1522 + ShouldNotReachHere(); 1.1523 + } 1.1524 +} 1.1525 + 1.1526 + 1.1527 +void TemplateTable::lcmp() { 1.1528 + transition(ltos, itos); 1.1529 + // y = rdx:rax 1.1530 + __ pop_l(rbx, rcx); // get x = rcx:rbx 1.1531 + __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y) 1.1532 + __ mov(rax, rcx); 1.1533 +} 1.1534 + 1.1535 + 1.1536 +void TemplateTable::float_cmp(bool is_float, int unordered_result) { 1.1537 + if (is_float) { 1.1538 + __ fld_s(at_rsp()); 1.1539 + } else { 1.1540 + __ fld_d(at_rsp()); 1.1541 + __ pop(rdx); 1.1542 + } 1.1543 + __ pop(rcx); 1.1544 + __ fcmp2int(rax, unordered_result < 0); 1.1545 +} 1.1546 + 1.1547 + 1.1548 +void TemplateTable::branch(bool is_jsr, bool is_wide) { 1.1549 + __ get_method(rcx); // ECX holds method 1.1550 + __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count 1.1551 + 1.1552 + const ByteSize be_offset = MethodCounters::backedge_counter_offset() + 1.1553 + InvocationCounter::counter_offset(); 1.1554 + const ByteSize inv_offset = MethodCounters::invocation_counter_offset() + 1.1555 + InvocationCounter::counter_offset(); 1.1556 + 1.1557 + // Load up EDX with the branch displacement 1.1558 + if (is_wide) { 1.1559 + __ movl(rdx, at_bcp(1)); 1.1560 + } else { 1.1561 + __ load_signed_short(rdx, at_bcp(1)); 1.1562 + } 1.1563 + __ bswapl(rdx); 1.1564 + if (!is_wide) __ sarl(rdx, 16); 1.1565 + LP64_ONLY(__ movslq(rdx, rdx)); 1.1566 + 1.1567 + 1.1568 + // Handle all the JSR stuff here, then exit. 1.1569 + // It's much shorter and cleaner than intermingling with the 1.1570 + // non-JSR normal-branch stuff occurring below. 1.1571 + if (is_jsr) { 1.1572 + // Pre-load the next target bytecode into EBX 1.1573 + __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0)); 1.1574 + 1.1575 + // compute return address as bci in rax, 1.1576 + __ lea(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(ConstMethod::codes_offset()))); 1.1577 + __ subptr(rax, Address(rcx, Method::const_offset())); 1.1578 + // Adjust the bcp in RSI by the displacement in EDX 1.1579 + __ addptr(rsi, rdx); 1.1580 + // Push return address 1.1581 + __ push_i(rax); 1.1582 + // jsr returns vtos 1.1583 + __ dispatch_only_noverify(vtos); 1.1584 + return; 1.1585 + } 1.1586 + 1.1587 + // Normal (non-jsr) branch handling 1.1588 + 1.1589 + // Adjust the bcp in RSI by the displacement in EDX 1.1590 + __ addptr(rsi, rdx); 1.1591 + 1.1592 + assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters"); 1.1593 + Label backedge_counter_overflow; 1.1594 + Label profile_method; 1.1595 + Label dispatch; 1.1596 + if (UseLoopCounter) { 1.1597 + // increment backedge counter for backward branches 1.1598 + // rax,: MDO 1.1599 + // rbx,: MDO bumped taken-count 1.1600 + // rcx: method 1.1601 + // rdx: target offset 1.1602 + // rsi: target bcp 1.1603 + // rdi: locals pointer 1.1604 + __ testl(rdx, rdx); // check if forward or backward branch 1.1605 + __ jcc(Assembler::positive, dispatch); // count only if backward branch 1.1606 + 1.1607 + // check if MethodCounters exists 1.1608 + Label has_counters; 1.1609 + __ movptr(rax, Address(rcx, Method::method_counters_offset())); 1.1610 + __ testptr(rax, rax); 1.1611 + __ jcc(Assembler::notZero, has_counters); 1.1612 + __ push(rdx); 1.1613 + __ push(rcx); 1.1614 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), 1.1615 + rcx); 1.1616 + __ pop(rcx); 1.1617 + __ pop(rdx); 1.1618 + __ movptr(rax, Address(rcx, Method::method_counters_offset())); 1.1619 + __ testptr(rax, rax); 1.1620 + __ jcc(Assembler::zero, dispatch); 1.1621 + __ bind(has_counters); 1.1622 + 1.1623 + if (TieredCompilation) { 1.1624 + Label no_mdo; 1.1625 + int increment = InvocationCounter::count_increment; 1.1626 + int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift; 1.1627 + if (ProfileInterpreter) { 1.1628 + // Are we profiling? 1.1629 + __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset()))); 1.1630 + __ testptr(rbx, rbx); 1.1631 + __ jccb(Assembler::zero, no_mdo); 1.1632 + // Increment the MDO backedge counter 1.1633 + const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) + 1.1634 + in_bytes(InvocationCounter::counter_offset())); 1.1635 + __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, 1.1636 + rax, false, Assembler::zero, &backedge_counter_overflow); 1.1637 + __ jmp(dispatch); 1.1638 + } 1.1639 + __ bind(no_mdo); 1.1640 + // Increment backedge counter in MethodCounters* 1.1641 + __ movptr(rcx, Address(rcx, Method::method_counters_offset())); 1.1642 + __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask, 1.1643 + rax, false, Assembler::zero, &backedge_counter_overflow); 1.1644 + } else { 1.1645 + // increment counter 1.1646 + __ movptr(rcx, Address(rcx, Method::method_counters_offset())); 1.1647 + __ movl(rax, Address(rcx, be_offset)); // load backedge counter 1.1648 + __ incrementl(rax, InvocationCounter::count_increment); // increment counter 1.1649 + __ movl(Address(rcx, be_offset), rax); // store counter 1.1650 + 1.1651 + __ movl(rax, Address(rcx, inv_offset)); // load invocation counter 1.1652 + 1.1653 + __ andl(rax, InvocationCounter::count_mask_value); // and the status bits 1.1654 + __ addl(rax, Address(rcx, be_offset)); // add both counters 1.1655 + 1.1656 + if (ProfileInterpreter) { 1.1657 + // Test to see if we should create a method data oop 1.1658 + __ cmp32(rax, 1.1659 + ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit)); 1.1660 + __ jcc(Assembler::less, dispatch); 1.1661 + 1.1662 + // if no method data exists, go to profile method 1.1663 + __ test_method_data_pointer(rax, profile_method); 1.1664 + 1.1665 + if (UseOnStackReplacement) { 1.1666 + // check for overflow against rbx, which is the MDO taken count 1.1667 + __ cmp32(rbx, 1.1668 + ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit)); 1.1669 + __ jcc(Assembler::below, dispatch); 1.1670 + 1.1671 + // When ProfileInterpreter is on, the backedge_count comes from the 1.1672 + // MethodData*, which value does not get reset on the call to 1.1673 + // frequency_counter_overflow(). To avoid excessive calls to the overflow 1.1674 + // routine while the method is being compiled, add a second test to make 1.1675 + // sure the overflow function is called only once every overflow_frequency. 1.1676 + const int overflow_frequency = 1024; 1.1677 + __ andptr(rbx, overflow_frequency-1); 1.1678 + __ jcc(Assembler::zero, backedge_counter_overflow); 1.1679 + } 1.1680 + } else { 1.1681 + if (UseOnStackReplacement) { 1.1682 + // check for overflow against rax, which is the sum of the counters 1.1683 + __ cmp32(rax, 1.1684 + ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit)); 1.1685 + __ jcc(Assembler::aboveEqual, backedge_counter_overflow); 1.1686 + 1.1687 + } 1.1688 + } 1.1689 + } 1.1690 + __ bind(dispatch); 1.1691 + } 1.1692 + 1.1693 + // Pre-load the next target bytecode into EBX 1.1694 + __ load_unsigned_byte(rbx, Address(rsi, 0)); 1.1695 + 1.1696 + // continue with the bytecode @ target 1.1697 + // rax,: return bci for jsr's, unused otherwise 1.1698 + // rbx,: target bytecode 1.1699 + // rsi: target bcp 1.1700 + __ dispatch_only(vtos); 1.1701 + 1.1702 + if (UseLoopCounter) { 1.1703 + if (ProfileInterpreter) { 1.1704 + // Out-of-line code to allocate method data oop. 1.1705 + __ bind(profile_method); 1.1706 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); 1.1707 + __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode 1.1708 + __ set_method_data_pointer_for_bcp(); 1.1709 + __ jmp(dispatch); 1.1710 + } 1.1711 + 1.1712 + if (UseOnStackReplacement) { 1.1713 + 1.1714 + // invocation counter overflow 1.1715 + __ bind(backedge_counter_overflow); 1.1716 + __ negptr(rdx); 1.1717 + __ addptr(rdx, rsi); // branch bcp 1.1718 + call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx); 1.1719 + __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode 1.1720 + 1.1721 + // rax,: osr nmethod (osr ok) or NULL (osr not possible) 1.1722 + // rbx,: target bytecode 1.1723 + // rdx: scratch 1.1724 + // rdi: locals pointer 1.1725 + // rsi: bcp 1.1726 + __ testptr(rax, rax); // test result 1.1727 + __ jcc(Assembler::zero, dispatch); // no osr if null 1.1728 + // nmethod may have been invalidated (VM may block upon call_VM return) 1.1729 + __ movl(rcx, Address(rax, nmethod::entry_bci_offset())); 1.1730 + __ cmpl(rcx, InvalidOSREntryBci); 1.1731 + __ jcc(Assembler::equal, dispatch); 1.1732 + 1.1733 + // We have the address of an on stack replacement routine in rax, 1.1734 + // We need to prepare to execute the OSR method. First we must 1.1735 + // migrate the locals and monitors off of the stack. 1.1736 + 1.1737 + __ mov(rbx, rax); // save the nmethod 1.1738 + 1.1739 + const Register thread = rcx; 1.1740 + __ get_thread(thread); 1.1741 + call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin)); 1.1742 + // rax, is OSR buffer, move it to expected parameter location 1.1743 + __ mov(rcx, rax); 1.1744 + 1.1745 + // pop the interpreter frame 1.1746 + __ movptr(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp 1.1747 + __ leave(); // remove frame anchor 1.1748 + __ pop(rdi); // get return address 1.1749 + __ mov(rsp, rdx); // set sp to sender sp 1.1750 + 1.1751 + // Align stack pointer for compiled code (note that caller is 1.1752 + // responsible for undoing this fixup by remembering the old SP 1.1753 + // in an rbp,-relative location) 1.1754 + __ andptr(rsp, -(StackAlignmentInBytes)); 1.1755 + 1.1756 + // push the (possibly adjusted) return address 1.1757 + __ push(rdi); 1.1758 + 1.1759 + // and begin the OSR nmethod 1.1760 + __ jmp(Address(rbx, nmethod::osr_entry_point_offset())); 1.1761 + } 1.1762 + } 1.1763 +} 1.1764 + 1.1765 + 1.1766 +void TemplateTable::if_0cmp(Condition cc) { 1.1767 + transition(itos, vtos); 1.1768 + // assume branch is more often taken than not (loops use backward branches) 1.1769 + Label not_taken; 1.1770 + __ testl(rax, rax); 1.1771 + __ jcc(j_not(cc), not_taken); 1.1772 + branch(false, false); 1.1773 + __ bind(not_taken); 1.1774 + __ profile_not_taken_branch(rax); 1.1775 +} 1.1776 + 1.1777 + 1.1778 +void TemplateTable::if_icmp(Condition cc) { 1.1779 + transition(itos, vtos); 1.1780 + // assume branch is more often taken than not (loops use backward branches) 1.1781 + Label not_taken; 1.1782 + __ pop_i(rdx); 1.1783 + __ cmpl(rdx, rax); 1.1784 + __ jcc(j_not(cc), not_taken); 1.1785 + branch(false, false); 1.1786 + __ bind(not_taken); 1.1787 + __ profile_not_taken_branch(rax); 1.1788 +} 1.1789 + 1.1790 + 1.1791 +void TemplateTable::if_nullcmp(Condition cc) { 1.1792 + transition(atos, vtos); 1.1793 + // assume branch is more often taken than not (loops use backward branches) 1.1794 + Label not_taken; 1.1795 + __ testptr(rax, rax); 1.1796 + __ jcc(j_not(cc), not_taken); 1.1797 + branch(false, false); 1.1798 + __ bind(not_taken); 1.1799 + __ profile_not_taken_branch(rax); 1.1800 +} 1.1801 + 1.1802 + 1.1803 +void TemplateTable::if_acmp(Condition cc) { 1.1804 + transition(atos, vtos); 1.1805 + // assume branch is more often taken than not (loops use backward branches) 1.1806 + Label not_taken; 1.1807 + __ pop_ptr(rdx); 1.1808 + __ cmpptr(rdx, rax); 1.1809 + __ jcc(j_not(cc), not_taken); 1.1810 + branch(false, false); 1.1811 + __ bind(not_taken); 1.1812 + __ profile_not_taken_branch(rax); 1.1813 +} 1.1814 + 1.1815 + 1.1816 +void TemplateTable::ret() { 1.1817 + transition(vtos, vtos); 1.1818 + locals_index(rbx); 1.1819 + __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp 1.1820 + __ profile_ret(rbx, rcx); 1.1821 + __ get_method(rax); 1.1822 + __ movptr(rsi, Address(rax, Method::const_offset())); 1.1823 + __ lea(rsi, Address(rsi, rbx, Address::times_1, 1.1824 + ConstMethod::codes_offset())); 1.1825 + __ dispatch_next(vtos); 1.1826 +} 1.1827 + 1.1828 + 1.1829 +void TemplateTable::wide_ret() { 1.1830 + transition(vtos, vtos); 1.1831 + locals_index_wide(rbx); 1.1832 + __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp 1.1833 + __ profile_ret(rbx, rcx); 1.1834 + __ get_method(rax); 1.1835 + __ movptr(rsi, Address(rax, Method::const_offset())); 1.1836 + __ lea(rsi, Address(rsi, rbx, Address::times_1, ConstMethod::codes_offset())); 1.1837 + __ dispatch_next(vtos); 1.1838 +} 1.1839 + 1.1840 + 1.1841 +void TemplateTable::tableswitch() { 1.1842 + Label default_case, continue_execution; 1.1843 + transition(itos, vtos); 1.1844 + // align rsi 1.1845 + __ lea(rbx, at_bcp(wordSize)); 1.1846 + __ andptr(rbx, -wordSize); 1.1847 + // load lo & hi 1.1848 + __ movl(rcx, Address(rbx, 1 * wordSize)); 1.1849 + __ movl(rdx, Address(rbx, 2 * wordSize)); 1.1850 + __ bswapl(rcx); 1.1851 + __ bswapl(rdx); 1.1852 + // check against lo & hi 1.1853 + __ cmpl(rax, rcx); 1.1854 + __ jccb(Assembler::less, default_case); 1.1855 + __ cmpl(rax, rdx); 1.1856 + __ jccb(Assembler::greater, default_case); 1.1857 + // lookup dispatch offset 1.1858 + __ subl(rax, rcx); 1.1859 + __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt)); 1.1860 + __ profile_switch_case(rax, rbx, rcx); 1.1861 + // continue execution 1.1862 + __ bind(continue_execution); 1.1863 + __ bswapl(rdx); 1.1864 + __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1)); 1.1865 + __ addptr(rsi, rdx); 1.1866 + __ dispatch_only(vtos); 1.1867 + // handle default 1.1868 + __ bind(default_case); 1.1869 + __ profile_switch_default(rax); 1.1870 + __ movl(rdx, Address(rbx, 0)); 1.1871 + __ jmp(continue_execution); 1.1872 +} 1.1873 + 1.1874 + 1.1875 +void TemplateTable::lookupswitch() { 1.1876 + transition(itos, itos); 1.1877 + __ stop("lookupswitch bytecode should have been rewritten"); 1.1878 +} 1.1879 + 1.1880 + 1.1881 +void TemplateTable::fast_linearswitch() { 1.1882 + transition(itos, vtos); 1.1883 + Label loop_entry, loop, found, continue_execution; 1.1884 + // bswapl rax, so we can avoid bswapping the table entries 1.1885 + __ bswapl(rax); 1.1886 + // align rsi 1.1887 + __ lea(rbx, at_bcp(wordSize)); // btw: should be able to get rid of this instruction (change offsets below) 1.1888 + __ andptr(rbx, -wordSize); 1.1889 + // set counter 1.1890 + __ movl(rcx, Address(rbx, wordSize)); 1.1891 + __ bswapl(rcx); 1.1892 + __ jmpb(loop_entry); 1.1893 + // table search 1.1894 + __ bind(loop); 1.1895 + __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize)); 1.1896 + __ jccb(Assembler::equal, found); 1.1897 + __ bind(loop_entry); 1.1898 + __ decrementl(rcx); 1.1899 + __ jcc(Assembler::greaterEqual, loop); 1.1900 + // default case 1.1901 + __ profile_switch_default(rax); 1.1902 + __ movl(rdx, Address(rbx, 0)); 1.1903 + __ jmpb(continue_execution); 1.1904 + // entry found -> get offset 1.1905 + __ bind(found); 1.1906 + __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize)); 1.1907 + __ profile_switch_case(rcx, rax, rbx); 1.1908 + // continue execution 1.1909 + __ bind(continue_execution); 1.1910 + __ bswapl(rdx); 1.1911 + __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1)); 1.1912 + __ addptr(rsi, rdx); 1.1913 + __ dispatch_only(vtos); 1.1914 +} 1.1915 + 1.1916 + 1.1917 +void TemplateTable::fast_binaryswitch() { 1.1918 + transition(itos, vtos); 1.1919 + // Implementation using the following core algorithm: 1.1920 + // 1.1921 + // int binary_search(int key, LookupswitchPair* array, int n) { 1.1922 + // // Binary search according to "Methodik des Programmierens" by 1.1923 + // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985. 1.1924 + // int i = 0; 1.1925 + // int j = n; 1.1926 + // while (i+1 < j) { 1.1927 + // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q) 1.1928 + // // with Q: for all i: 0 <= i < n: key < a[i] 1.1929 + // // where a stands for the array and assuming that the (inexisting) 1.1930 + // // element a[n] is infinitely big. 1.1931 + // int h = (i + j) >> 1; 1.1932 + // // i < h < j 1.1933 + // if (key < array[h].fast_match()) { 1.1934 + // j = h; 1.1935 + // } else { 1.1936 + // i = h; 1.1937 + // } 1.1938 + // } 1.1939 + // // R: a[i] <= key < a[i+1] or Q 1.1940 + // // (i.e., if key is within array, i is the correct index) 1.1941 + // return i; 1.1942 + // } 1.1943 + 1.1944 + // register allocation 1.1945 + const Register key = rax; // already set (tosca) 1.1946 + const Register array = rbx; 1.1947 + const Register i = rcx; 1.1948 + const Register j = rdx; 1.1949 + const Register h = rdi; // needs to be restored 1.1950 + const Register temp = rsi; 1.1951 + // setup array 1.1952 + __ save_bcp(); 1.1953 + 1.1954 + __ lea(array, at_bcp(3*wordSize)); // btw: should be able to get rid of this instruction (change offsets below) 1.1955 + __ andptr(array, -wordSize); 1.1956 + // initialize i & j 1.1957 + __ xorl(i, i); // i = 0; 1.1958 + __ movl(j, Address(array, -wordSize)); // j = length(array); 1.1959 + // Convert j into native byteordering 1.1960 + __ bswapl(j); 1.1961 + // and start 1.1962 + Label entry; 1.1963 + __ jmp(entry); 1.1964 + 1.1965 + // binary search loop 1.1966 + { Label loop; 1.1967 + __ bind(loop); 1.1968 + // int h = (i + j) >> 1; 1.1969 + __ leal(h, Address(i, j, Address::times_1)); // h = i + j; 1.1970 + __ sarl(h, 1); // h = (i + j) >> 1; 1.1971 + // if (key < array[h].fast_match()) { 1.1972 + // j = h; 1.1973 + // } else { 1.1974 + // i = h; 1.1975 + // } 1.1976 + // Convert array[h].match to native byte-ordering before compare 1.1977 + __ movl(temp, Address(array, h, Address::times_8, 0*wordSize)); 1.1978 + __ bswapl(temp); 1.1979 + __ cmpl(key, temp); 1.1980 + // j = h if (key < array[h].fast_match()) 1.1981 + __ cmov32(Assembler::less , j, h); 1.1982 + // i = h if (key >= array[h].fast_match()) 1.1983 + __ cmov32(Assembler::greaterEqual, i, h); 1.1984 + // while (i+1 < j) 1.1985 + __ bind(entry); 1.1986 + __ leal(h, Address(i, 1)); // i+1 1.1987 + __ cmpl(h, j); // i+1 < j 1.1988 + __ jcc(Assembler::less, loop); 1.1989 + } 1.1990 + 1.1991 + // end of binary search, result index is i (must check again!) 1.1992 + Label default_case; 1.1993 + // Convert array[i].match to native byte-ordering before compare 1.1994 + __ movl(temp, Address(array, i, Address::times_8, 0*wordSize)); 1.1995 + __ bswapl(temp); 1.1996 + __ cmpl(key, temp); 1.1997 + __ jcc(Assembler::notEqual, default_case); 1.1998 + 1.1999 + // entry found -> j = offset 1.2000 + __ movl(j , Address(array, i, Address::times_8, 1*wordSize)); 1.2001 + __ profile_switch_case(i, key, array); 1.2002 + __ bswapl(j); 1.2003 + LP64_ONLY(__ movslq(j, j)); 1.2004 + __ restore_bcp(); 1.2005 + __ restore_locals(); // restore rdi 1.2006 + __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1)); 1.2007 + 1.2008 + __ addptr(rsi, j); 1.2009 + __ dispatch_only(vtos); 1.2010 + 1.2011 + // default case -> j = default offset 1.2012 + __ bind(default_case); 1.2013 + __ profile_switch_default(i); 1.2014 + __ movl(j, Address(array, -2*wordSize)); 1.2015 + __ bswapl(j); 1.2016 + LP64_ONLY(__ movslq(j, j)); 1.2017 + __ restore_bcp(); 1.2018 + __ restore_locals(); // restore rdi 1.2019 + __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1)); 1.2020 + __ addptr(rsi, j); 1.2021 + __ dispatch_only(vtos); 1.2022 +} 1.2023 + 1.2024 + 1.2025 +void TemplateTable::_return(TosState state) { 1.2026 + transition(state, state); 1.2027 + assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation 1.2028 + 1.2029 + if (_desc->bytecode() == Bytecodes::_return_register_finalizer) { 1.2030 + assert(state == vtos, "only valid state"); 1.2031 + __ movptr(rax, aaddress(0)); 1.2032 + __ load_klass(rdi, rax); 1.2033 + __ movl(rdi, Address(rdi, Klass::access_flags_offset())); 1.2034 + __ testl(rdi, JVM_ACC_HAS_FINALIZER); 1.2035 + Label skip_register_finalizer; 1.2036 + __ jcc(Assembler::zero, skip_register_finalizer); 1.2037 + 1.2038 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax); 1.2039 + 1.2040 + __ bind(skip_register_finalizer); 1.2041 + } 1.2042 + 1.2043 + __ remove_activation(state, rsi); 1.2044 + __ jmp(rsi); 1.2045 +} 1.2046 + 1.2047 + 1.2048 +// ---------------------------------------------------------------------------- 1.2049 +// Volatile variables demand their effects be made known to all CPU's in 1.2050 +// order. Store buffers on most chips allow reads & writes to reorder; the 1.2051 +// JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of 1.2052 +// memory barrier (i.e., it's not sufficient that the interpreter does not 1.2053 +// reorder volatile references, the hardware also must not reorder them). 1.2054 +// 1.2055 +// According to the new Java Memory Model (JMM): 1.2056 +// (1) All volatiles are serialized wrt to each other. 1.2057 +// ALSO reads & writes act as aquire & release, so: 1.2058 +// (2) A read cannot let unrelated NON-volatile memory refs that happen after 1.2059 +// the read float up to before the read. It's OK for non-volatile memory refs 1.2060 +// that happen before the volatile read to float down below it. 1.2061 +// (3) Similar a volatile write cannot let unrelated NON-volatile memory refs 1.2062 +// that happen BEFORE the write float down to after the write. It's OK for 1.2063 +// non-volatile memory refs that happen after the volatile write to float up 1.2064 +// before it. 1.2065 +// 1.2066 +// We only put in barriers around volatile refs (they are expensive), not 1.2067 +// _between_ memory refs (that would require us to track the flavor of the 1.2068 +// previous memory refs). Requirements (2) and (3) require some barriers 1.2069 +// before volatile stores and after volatile loads. These nearly cover 1.2070 +// requirement (1) but miss the volatile-store-volatile-load case. This final 1.2071 +// case is placed after volatile-stores although it could just as well go 1.2072 +// before volatile-loads. 1.2073 +void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) { 1.2074 + // Helper function to insert a is-volatile test and memory barrier 1.2075 + if( !os::is_MP() ) return; // Not needed on single CPU 1.2076 + __ membar(order_constraint); 1.2077 +} 1.2078 + 1.2079 +void TemplateTable::resolve_cache_and_index(int byte_no, 1.2080 + Register Rcache, 1.2081 + Register index, 1.2082 + size_t index_size) { 1.2083 + const Register temp = rbx; 1.2084 + assert_different_registers(Rcache, index, temp); 1.2085 + 1.2086 + Label resolved; 1.2087 + assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range"); 1.2088 + __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size); 1.2089 + __ cmpl(temp, (int) bytecode()); // have we resolved this bytecode? 1.2090 + __ jcc(Assembler::equal, resolved); 1.2091 + 1.2092 + // resolve first time through 1.2093 + address entry; 1.2094 + switch (bytecode()) { 1.2095 + case Bytecodes::_getstatic : // fall through 1.2096 + case Bytecodes::_putstatic : // fall through 1.2097 + case Bytecodes::_getfield : // fall through 1.2098 + case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break; 1.2099 + case Bytecodes::_invokevirtual : // fall through 1.2100 + case Bytecodes::_invokespecial : // fall through 1.2101 + case Bytecodes::_invokestatic : // fall through 1.2102 + case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break; 1.2103 + case Bytecodes::_invokehandle : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle); break; 1.2104 + case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break; 1.2105 + default: 1.2106 + fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode()))); 1.2107 + break; 1.2108 + } 1.2109 + __ movl(temp, (int)bytecode()); 1.2110 + __ call_VM(noreg, entry, temp); 1.2111 + // Update registers with resolved info 1.2112 + __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size); 1.2113 + __ bind(resolved); 1.2114 +} 1.2115 + 1.2116 + 1.2117 +// The cache and index registers must be set before call 1.2118 +void TemplateTable::load_field_cp_cache_entry(Register obj, 1.2119 + Register cache, 1.2120 + Register index, 1.2121 + Register off, 1.2122 + Register flags, 1.2123 + bool is_static = false) { 1.2124 + assert_different_registers(cache, index, flags, off); 1.2125 + 1.2126 + ByteSize cp_base_offset = ConstantPoolCache::base_offset(); 1.2127 + // Field offset 1.2128 + __ movptr(off, Address(cache, index, Address::times_ptr, 1.2129 + in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset()))); 1.2130 + // Flags 1.2131 + __ movl(flags, Address(cache, index, Address::times_ptr, 1.2132 + in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset()))); 1.2133 + 1.2134 + // klass overwrite register 1.2135 + if (is_static) { 1.2136 + __ movptr(obj, Address(cache, index, Address::times_ptr, 1.2137 + in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset()))); 1.2138 + const int mirror_offset = in_bytes(Klass::java_mirror_offset()); 1.2139 + __ movptr(obj, Address(obj, mirror_offset)); 1.2140 + } 1.2141 +} 1.2142 + 1.2143 +void TemplateTable::load_invoke_cp_cache_entry(int byte_no, 1.2144 + Register method, 1.2145 + Register itable_index, 1.2146 + Register flags, 1.2147 + bool is_invokevirtual, 1.2148 + bool is_invokevfinal, /*unused*/ 1.2149 + bool is_invokedynamic) { 1.2150 + // setup registers 1.2151 + const Register cache = rcx; 1.2152 + const Register index = rdx; 1.2153 + assert_different_registers(method, flags); 1.2154 + assert_different_registers(method, cache, index); 1.2155 + assert_different_registers(itable_index, flags); 1.2156 + assert_different_registers(itable_index, cache, index); 1.2157 + // determine constant pool cache field offsets 1.2158 + assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant"); 1.2159 + const int method_offset = in_bytes( 1.2160 + ConstantPoolCache::base_offset() + 1.2161 + ((byte_no == f2_byte) 1.2162 + ? ConstantPoolCacheEntry::f2_offset() 1.2163 + : ConstantPoolCacheEntry::f1_offset())); 1.2164 + const int flags_offset = in_bytes(ConstantPoolCache::base_offset() + 1.2165 + ConstantPoolCacheEntry::flags_offset()); 1.2166 + // access constant pool cache fields 1.2167 + const int index_offset = in_bytes(ConstantPoolCache::base_offset() + 1.2168 + ConstantPoolCacheEntry::f2_offset()); 1.2169 + 1.2170 + size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2)); 1.2171 + resolve_cache_and_index(byte_no, cache, index, index_size); 1.2172 + __ movptr(method, Address(cache, index, Address::times_ptr, method_offset)); 1.2173 + 1.2174 + if (itable_index != noreg) { 1.2175 + __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset)); 1.2176 + } 1.2177 + __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset)); 1.2178 +} 1.2179 + 1.2180 + 1.2181 +// The registers cache and index expected to be set before call. 1.2182 +// Correct values of the cache and index registers are preserved. 1.2183 +void TemplateTable::jvmti_post_field_access(Register cache, 1.2184 + Register index, 1.2185 + bool is_static, 1.2186 + bool has_tos) { 1.2187 + if (JvmtiExport::can_post_field_access()) { 1.2188 + // Check to see if a field access watch has been set before we take 1.2189 + // the time to call into the VM. 1.2190 + Label L1; 1.2191 + assert_different_registers(cache, index, rax); 1.2192 + __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr())); 1.2193 + __ testl(rax,rax); 1.2194 + __ jcc(Assembler::zero, L1); 1.2195 + 1.2196 + // cache entry pointer 1.2197 + __ addptr(cache, in_bytes(ConstantPoolCache::base_offset())); 1.2198 + __ shll(index, LogBytesPerWord); 1.2199 + __ addptr(cache, index); 1.2200 + if (is_static) { 1.2201 + __ xorptr(rax, rax); // NULL object reference 1.2202 + } else { 1.2203 + __ pop(atos); // Get the object 1.2204 + __ verify_oop(rax); 1.2205 + __ push(atos); // Restore stack state 1.2206 + } 1.2207 + // rax,: object pointer or NULL 1.2208 + // cache: cache entry pointer 1.2209 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), 1.2210 + rax, cache); 1.2211 + __ get_cache_and_index_at_bcp(cache, index, 1); 1.2212 + __ bind(L1); 1.2213 + } 1.2214 +} 1.2215 + 1.2216 +void TemplateTable::pop_and_check_object(Register r) { 1.2217 + __ pop_ptr(r); 1.2218 + __ null_check(r); // for field access must check obj. 1.2219 + __ verify_oop(r); 1.2220 +} 1.2221 + 1.2222 +void TemplateTable::getfield_or_static(int byte_no, bool is_static) { 1.2223 + transition(vtos, vtos); 1.2224 + 1.2225 + const Register cache = rcx; 1.2226 + const Register index = rdx; 1.2227 + const Register obj = rcx; 1.2228 + const Register off = rbx; 1.2229 + const Register flags = rax; 1.2230 + 1.2231 + resolve_cache_and_index(byte_no, cache, index, sizeof(u2)); 1.2232 + jvmti_post_field_access(cache, index, is_static, false); 1.2233 + load_field_cp_cache_entry(obj, cache, index, off, flags, is_static); 1.2234 + 1.2235 + if (!is_static) pop_and_check_object(obj); 1.2236 + 1.2237 + const Address lo(obj, off, Address::times_1, 0*wordSize); 1.2238 + const Address hi(obj, off, Address::times_1, 1*wordSize); 1.2239 + 1.2240 + Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble; 1.2241 + 1.2242 + __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift); 1.2243 + assert(btos == 0, "change code, btos != 0"); 1.2244 + // btos 1.2245 + __ andptr(flags, ConstantPoolCacheEntry::tos_state_mask); 1.2246 + __ jcc(Assembler::notZero, notByte); 1.2247 + 1.2248 + __ load_signed_byte(rax, lo ); 1.2249 + __ push(btos); 1.2250 + // Rewrite bytecode to be faster 1.2251 + if (!is_static) { 1.2252 + patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx); 1.2253 + } 1.2254 + __ jmp(Done); 1.2255 + 1.2256 + __ bind(notByte); 1.2257 + // itos 1.2258 + __ cmpl(flags, itos ); 1.2259 + __ jcc(Assembler::notEqual, notInt); 1.2260 + 1.2261 + __ movl(rax, lo ); 1.2262 + __ push(itos); 1.2263 + // Rewrite bytecode to be faster 1.2264 + if (!is_static) { 1.2265 + patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx); 1.2266 + } 1.2267 + __ jmp(Done); 1.2268 + 1.2269 + __ bind(notInt); 1.2270 + // atos 1.2271 + __ cmpl(flags, atos ); 1.2272 + __ jcc(Assembler::notEqual, notObj); 1.2273 + 1.2274 + __ movl(rax, lo ); 1.2275 + __ push(atos); 1.2276 + if (!is_static) { 1.2277 + patch_bytecode(Bytecodes::_fast_agetfield, rcx, rbx); 1.2278 + } 1.2279 + __ jmp(Done); 1.2280 + 1.2281 + __ bind(notObj); 1.2282 + // ctos 1.2283 + __ cmpl(flags, ctos ); 1.2284 + __ jcc(Assembler::notEqual, notChar); 1.2285 + 1.2286 + __ load_unsigned_short(rax, lo ); 1.2287 + __ push(ctos); 1.2288 + if (!is_static) { 1.2289 + patch_bytecode(Bytecodes::_fast_cgetfield, rcx, rbx); 1.2290 + } 1.2291 + __ jmp(Done); 1.2292 + 1.2293 + __ bind(notChar); 1.2294 + // stos 1.2295 + __ cmpl(flags, stos ); 1.2296 + __ jcc(Assembler::notEqual, notShort); 1.2297 + 1.2298 + __ load_signed_short(rax, lo ); 1.2299 + __ push(stos); 1.2300 + if (!is_static) { 1.2301 + patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx); 1.2302 + } 1.2303 + __ jmp(Done); 1.2304 + 1.2305 + __ bind(notShort); 1.2306 + // ltos 1.2307 + __ cmpl(flags, ltos ); 1.2308 + __ jcc(Assembler::notEqual, notLong); 1.2309 + 1.2310 + // Generate code as if volatile. There just aren't enough registers to 1.2311 + // save that information and this code is faster than the test. 1.2312 + __ fild_d(lo); // Must load atomically 1.2313 + __ subptr(rsp,2*wordSize); // Make space for store 1.2314 + __ fistp_d(Address(rsp,0)); 1.2315 + __ pop(rax); 1.2316 + __ pop(rdx); 1.2317 + 1.2318 + __ push(ltos); 1.2319 + // Don't rewrite to _fast_lgetfield for potential volatile case. 1.2320 + __ jmp(Done); 1.2321 + 1.2322 + __ bind(notLong); 1.2323 + // ftos 1.2324 + __ cmpl(flags, ftos ); 1.2325 + __ jcc(Assembler::notEqual, notFloat); 1.2326 + 1.2327 + __ fld_s(lo); 1.2328 + __ push(ftos); 1.2329 + if (!is_static) { 1.2330 + patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx); 1.2331 + } 1.2332 + __ jmp(Done); 1.2333 + 1.2334 + __ bind(notFloat); 1.2335 + // dtos 1.2336 + __ cmpl(flags, dtos ); 1.2337 + __ jcc(Assembler::notEqual, notDouble); 1.2338 + 1.2339 + __ fld_d(lo); 1.2340 + __ push(dtos); 1.2341 + if (!is_static) { 1.2342 + patch_bytecode(Bytecodes::_fast_dgetfield, rcx, rbx); 1.2343 + } 1.2344 + __ jmpb(Done); 1.2345 + 1.2346 + __ bind(notDouble); 1.2347 + 1.2348 + __ stop("Bad state"); 1.2349 + 1.2350 + __ bind(Done); 1.2351 + // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO). 1.2352 + // volatile_barrier( ); 1.2353 +} 1.2354 + 1.2355 + 1.2356 +void TemplateTable::getfield(int byte_no) { 1.2357 + getfield_or_static(byte_no, false); 1.2358 +} 1.2359 + 1.2360 + 1.2361 +void TemplateTable::getstatic(int byte_no) { 1.2362 + getfield_or_static(byte_no, true); 1.2363 +} 1.2364 + 1.2365 +// The registers cache and index expected to be set before call. 1.2366 +// The function may destroy various registers, just not the cache and index registers. 1.2367 +void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) { 1.2368 + 1.2369 + ByteSize cp_base_offset = ConstantPoolCache::base_offset(); 1.2370 + 1.2371 + if (JvmtiExport::can_post_field_modification()) { 1.2372 + // Check to see if a field modification watch has been set before we take 1.2373 + // the time to call into the VM. 1.2374 + Label L1; 1.2375 + assert_different_registers(cache, index, rax); 1.2376 + __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr())); 1.2377 + __ testl(rax, rax); 1.2378 + __ jcc(Assembler::zero, L1); 1.2379 + 1.2380 + // The cache and index registers have been already set. 1.2381 + // This allows to eliminate this call but the cache and index 1.2382 + // registers have to be correspondingly used after this line. 1.2383 + __ get_cache_and_index_at_bcp(rax, rdx, 1); 1.2384 + 1.2385 + if (is_static) { 1.2386 + // Life is simple. Null out the object pointer. 1.2387 + __ xorptr(rbx, rbx); 1.2388 + } else { 1.2389 + // Life is harder. The stack holds the value on top, followed by the object. 1.2390 + // We don't know the size of the value, though; it could be one or two words 1.2391 + // depending on its type. As a result, we must find the type to determine where 1.2392 + // the object is. 1.2393 + Label two_word, valsize_known; 1.2394 + __ movl(rcx, Address(rax, rdx, Address::times_ptr, in_bytes(cp_base_offset + 1.2395 + ConstantPoolCacheEntry::flags_offset()))); 1.2396 + __ mov(rbx, rsp); 1.2397 + __ shrl(rcx, ConstantPoolCacheEntry::tos_state_shift); 1.2398 + // Make sure we don't need to mask rcx after the above shift 1.2399 + ConstantPoolCacheEntry::verify_tos_state_shift(); 1.2400 + __ cmpl(rcx, ltos); 1.2401 + __ jccb(Assembler::equal, two_word); 1.2402 + __ cmpl(rcx, dtos); 1.2403 + __ jccb(Assembler::equal, two_word); 1.2404 + __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos) 1.2405 + __ jmpb(valsize_known); 1.2406 + 1.2407 + __ bind(two_word); 1.2408 + __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue 1.2409 + 1.2410 + __ bind(valsize_known); 1.2411 + // setup object pointer 1.2412 + __ movptr(rbx, Address(rbx, 0)); 1.2413 + } 1.2414 + // cache entry pointer 1.2415 + __ addptr(rax, in_bytes(cp_base_offset)); 1.2416 + __ shll(rdx, LogBytesPerWord); 1.2417 + __ addptr(rax, rdx); 1.2418 + // object (tos) 1.2419 + __ mov(rcx, rsp); 1.2420 + // rbx,: object pointer set up above (NULL if static) 1.2421 + // rax,: cache entry pointer 1.2422 + // rcx: jvalue object on the stack 1.2423 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), 1.2424 + rbx, rax, rcx); 1.2425 + __ get_cache_and_index_at_bcp(cache, index, 1); 1.2426 + __ bind(L1); 1.2427 + } 1.2428 +} 1.2429 + 1.2430 + 1.2431 +void TemplateTable::putfield_or_static(int byte_no, bool is_static) { 1.2432 + transition(vtos, vtos); 1.2433 + 1.2434 + const Register cache = rcx; 1.2435 + const Register index = rdx; 1.2436 + const Register obj = rcx; 1.2437 + const Register off = rbx; 1.2438 + const Register flags = rax; 1.2439 + 1.2440 + resolve_cache_and_index(byte_no, cache, index, sizeof(u2)); 1.2441 + jvmti_post_field_mod(cache, index, is_static); 1.2442 + load_field_cp_cache_entry(obj, cache, index, off, flags, is_static); 1.2443 + 1.2444 + // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO). 1.2445 + // volatile_barrier( ); 1.2446 + 1.2447 + Label notVolatile, Done; 1.2448 + __ movl(rdx, flags); 1.2449 + __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift); 1.2450 + __ andl(rdx, 0x1); 1.2451 + 1.2452 + // field addresses 1.2453 + const Address lo(obj, off, Address::times_1, 0*wordSize); 1.2454 + const Address hi(obj, off, Address::times_1, 1*wordSize); 1.2455 + 1.2456 + Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble; 1.2457 + 1.2458 + __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift); 1.2459 + assert(btos == 0, "change code, btos != 0"); 1.2460 + __ andl(flags, ConstantPoolCacheEntry::tos_state_mask); 1.2461 + __ jcc(Assembler::notZero, notByte); 1.2462 + 1.2463 + // btos 1.2464 + { 1.2465 + __ pop(btos); 1.2466 + if (!is_static) pop_and_check_object(obj); 1.2467 + __ movb(lo, rax); 1.2468 + if (!is_static) { 1.2469 + patch_bytecode(Bytecodes::_fast_bputfield, rcx, rbx, true, byte_no); 1.2470 + } 1.2471 + __ jmp(Done); 1.2472 + } 1.2473 + 1.2474 + __ bind(notByte); 1.2475 + __ cmpl(flags, itos); 1.2476 + __ jcc(Assembler::notEqual, notInt); 1.2477 + 1.2478 + // itos 1.2479 + { 1.2480 + __ pop(itos); 1.2481 + if (!is_static) pop_and_check_object(obj); 1.2482 + __ movl(lo, rax); 1.2483 + if (!is_static) { 1.2484 + patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx, true, byte_no); 1.2485 + } 1.2486 + __ jmp(Done); 1.2487 + } 1.2488 + 1.2489 + __ bind(notInt); 1.2490 + __ cmpl(flags, atos); 1.2491 + __ jcc(Assembler::notEqual, notObj); 1.2492 + 1.2493 + // atos 1.2494 + { 1.2495 + __ pop(atos); 1.2496 + if (!is_static) pop_and_check_object(obj); 1.2497 + do_oop_store(_masm, lo, rax, _bs->kind(), false); 1.2498 + if (!is_static) { 1.2499 + patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx, true, byte_no); 1.2500 + } 1.2501 + __ jmp(Done); 1.2502 + } 1.2503 + 1.2504 + __ bind(notObj); 1.2505 + __ cmpl(flags, ctos); 1.2506 + __ jcc(Assembler::notEqual, notChar); 1.2507 + 1.2508 + // ctos 1.2509 + { 1.2510 + __ pop(ctos); 1.2511 + if (!is_static) pop_and_check_object(obj); 1.2512 + __ movw(lo, rax); 1.2513 + if (!is_static) { 1.2514 + patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx, true, byte_no); 1.2515 + } 1.2516 + __ jmp(Done); 1.2517 + } 1.2518 + 1.2519 + __ bind(notChar); 1.2520 + __ cmpl(flags, stos); 1.2521 + __ jcc(Assembler::notEqual, notShort); 1.2522 + 1.2523 + // stos 1.2524 + { 1.2525 + __ pop(stos); 1.2526 + if (!is_static) pop_and_check_object(obj); 1.2527 + __ movw(lo, rax); 1.2528 + if (!is_static) { 1.2529 + patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx, true, byte_no); 1.2530 + } 1.2531 + __ jmp(Done); 1.2532 + } 1.2533 + 1.2534 + __ bind(notShort); 1.2535 + __ cmpl(flags, ltos); 1.2536 + __ jcc(Assembler::notEqual, notLong); 1.2537 + 1.2538 + // ltos 1.2539 + { 1.2540 + Label notVolatileLong; 1.2541 + __ testl(rdx, rdx); 1.2542 + __ jcc(Assembler::zero, notVolatileLong); 1.2543 + 1.2544 + __ pop(ltos); // overwrites rdx, do this after testing volatile. 1.2545 + if (!is_static) pop_and_check_object(obj); 1.2546 + 1.2547 + // Replace with real volatile test 1.2548 + __ push(rdx); 1.2549 + __ push(rax); // Must update atomically with FIST 1.2550 + __ fild_d(Address(rsp,0)); // So load into FPU register 1.2551 + __ fistp_d(lo); // and put into memory atomically 1.2552 + __ addptr(rsp, 2*wordSize); 1.2553 + // volatile_barrier(); 1.2554 + volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 1.2555 + Assembler::StoreStore)); 1.2556 + // Don't rewrite volatile version 1.2557 + __ jmp(notVolatile); 1.2558 + 1.2559 + __ bind(notVolatileLong); 1.2560 + 1.2561 + __ pop(ltos); // overwrites rdx 1.2562 + if (!is_static) pop_and_check_object(obj); 1.2563 + NOT_LP64(__ movptr(hi, rdx)); 1.2564 + __ movptr(lo, rax); 1.2565 + if (!is_static) { 1.2566 + patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx, true, byte_no); 1.2567 + } 1.2568 + __ jmp(notVolatile); 1.2569 + } 1.2570 + 1.2571 + __ bind(notLong); 1.2572 + __ cmpl(flags, ftos); 1.2573 + __ jcc(Assembler::notEqual, notFloat); 1.2574 + 1.2575 + // ftos 1.2576 + { 1.2577 + __ pop(ftos); 1.2578 + if (!is_static) pop_and_check_object(obj); 1.2579 + __ fstp_s(lo); 1.2580 + if (!is_static) { 1.2581 + patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx, true, byte_no); 1.2582 + } 1.2583 + __ jmp(Done); 1.2584 + } 1.2585 + 1.2586 + __ bind(notFloat); 1.2587 +#ifdef ASSERT 1.2588 + __ cmpl(flags, dtos); 1.2589 + __ jcc(Assembler::notEqual, notDouble); 1.2590 +#endif 1.2591 + 1.2592 + // dtos 1.2593 + { 1.2594 + __ pop(dtos); 1.2595 + if (!is_static) pop_and_check_object(obj); 1.2596 + __ fstp_d(lo); 1.2597 + if (!is_static) { 1.2598 + patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx, true, byte_no); 1.2599 + } 1.2600 + __ jmp(Done); 1.2601 + } 1.2602 + 1.2603 +#ifdef ASSERT 1.2604 + __ bind(notDouble); 1.2605 + __ stop("Bad state"); 1.2606 +#endif 1.2607 + 1.2608 + __ bind(Done); 1.2609 + 1.2610 + // Check for volatile store 1.2611 + __ testl(rdx, rdx); 1.2612 + __ jcc(Assembler::zero, notVolatile); 1.2613 + volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 1.2614 + Assembler::StoreStore)); 1.2615 + __ bind(notVolatile); 1.2616 +} 1.2617 + 1.2618 + 1.2619 +void TemplateTable::putfield(int byte_no) { 1.2620 + putfield_or_static(byte_no, false); 1.2621 +} 1.2622 + 1.2623 + 1.2624 +void TemplateTable::putstatic(int byte_no) { 1.2625 + putfield_or_static(byte_no, true); 1.2626 +} 1.2627 + 1.2628 +void TemplateTable::jvmti_post_fast_field_mod() { 1.2629 + if (JvmtiExport::can_post_field_modification()) { 1.2630 + // Check to see if a field modification watch has been set before we take 1.2631 + // the time to call into the VM. 1.2632 + Label L2; 1.2633 + __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr())); 1.2634 + __ testl(rcx,rcx); 1.2635 + __ jcc(Assembler::zero, L2); 1.2636 + __ pop_ptr(rbx); // copy the object pointer from tos 1.2637 + __ verify_oop(rbx); 1.2638 + __ push_ptr(rbx); // put the object pointer back on tos 1.2639 + 1.2640 + // Save tos values before call_VM() clobbers them. Since we have 1.2641 + // to do it for every data type, we use the saved values as the 1.2642 + // jvalue object. 1.2643 + switch (bytecode()) { // load values into the jvalue object 1.2644 + case Bytecodes::_fast_aputfield: __ push_ptr(rax); break; 1.2645 + case Bytecodes::_fast_bputfield: // fall through 1.2646 + case Bytecodes::_fast_sputfield: // fall through 1.2647 + case Bytecodes::_fast_cputfield: // fall through 1.2648 + case Bytecodes::_fast_iputfield: __ push_i(rax); break; 1.2649 + case Bytecodes::_fast_dputfield: __ push_d(); break; 1.2650 + case Bytecodes::_fast_fputfield: __ push_f(); break; 1.2651 + case Bytecodes::_fast_lputfield: __ push_l(rax); break; 1.2652 + 1.2653 + default: 1.2654 + ShouldNotReachHere(); 1.2655 + } 1.2656 + __ mov(rcx, rsp); // points to jvalue on the stack 1.2657 + // access constant pool cache entry 1.2658 + __ get_cache_entry_pointer_at_bcp(rax, rdx, 1); 1.2659 + __ verify_oop(rbx); 1.2660 + // rbx,: object pointer copied above 1.2661 + // rax,: cache entry pointer 1.2662 + // rcx: jvalue object on the stack 1.2663 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx); 1.2664 + 1.2665 + switch (bytecode()) { // restore tos values 1.2666 + case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break; 1.2667 + case Bytecodes::_fast_bputfield: // fall through 1.2668 + case Bytecodes::_fast_sputfield: // fall through 1.2669 + case Bytecodes::_fast_cputfield: // fall through 1.2670 + case Bytecodes::_fast_iputfield: __ pop_i(rax); break; 1.2671 + case Bytecodes::_fast_dputfield: __ pop_d(); break; 1.2672 + case Bytecodes::_fast_fputfield: __ pop_f(); break; 1.2673 + case Bytecodes::_fast_lputfield: __ pop_l(rax); break; 1.2674 + } 1.2675 + __ bind(L2); 1.2676 + } 1.2677 +} 1.2678 + 1.2679 +void TemplateTable::fast_storefield(TosState state) { 1.2680 + transition(state, vtos); 1.2681 + 1.2682 + ByteSize base = ConstantPoolCache::base_offset(); 1.2683 + 1.2684 + jvmti_post_fast_field_mod(); 1.2685 + 1.2686 + // access constant pool cache 1.2687 + __ get_cache_and_index_at_bcp(rcx, rbx, 1); 1.2688 + 1.2689 + // test for volatile with rdx but rdx is tos register for lputfield. 1.2690 + if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx); 1.2691 + __ movl(rdx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + 1.2692 + ConstantPoolCacheEntry::flags_offset()))); 1.2693 + 1.2694 + // replace index with field offset from cache entry 1.2695 + __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + ConstantPoolCacheEntry::f2_offset()))); 1.2696 + 1.2697 + // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO). 1.2698 + // volatile_barrier( ); 1.2699 + 1.2700 + Label notVolatile, Done; 1.2701 + __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift); 1.2702 + __ andl(rdx, 0x1); 1.2703 + // Check for volatile store 1.2704 + __ testl(rdx, rdx); 1.2705 + __ jcc(Assembler::zero, notVolatile); 1.2706 + 1.2707 + if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx); 1.2708 + 1.2709 + // Get object from stack 1.2710 + pop_and_check_object(rcx); 1.2711 + 1.2712 + // field addresses 1.2713 + const Address lo(rcx, rbx, Address::times_1, 0*wordSize); 1.2714 + const Address hi(rcx, rbx, Address::times_1, 1*wordSize); 1.2715 + 1.2716 + // access field 1.2717 + switch (bytecode()) { 1.2718 + case Bytecodes::_fast_bputfield: __ movb(lo, rax); break; 1.2719 + case Bytecodes::_fast_sputfield: // fall through 1.2720 + case Bytecodes::_fast_cputfield: __ movw(lo, rax); break; 1.2721 + case Bytecodes::_fast_iputfield: __ movl(lo, rax); break; 1.2722 + case Bytecodes::_fast_lputfield: 1.2723 + NOT_LP64(__ movptr(hi, rdx)); 1.2724 + __ movptr(lo, rax); 1.2725 + break; 1.2726 + case Bytecodes::_fast_fputfield: __ fstp_s(lo); break; 1.2727 + case Bytecodes::_fast_dputfield: __ fstp_d(lo); break; 1.2728 + case Bytecodes::_fast_aputfield: { 1.2729 + do_oop_store(_masm, lo, rax, _bs->kind(), false); 1.2730 + break; 1.2731 + } 1.2732 + default: 1.2733 + ShouldNotReachHere(); 1.2734 + } 1.2735 + 1.2736 + Label done; 1.2737 + volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad | 1.2738 + Assembler::StoreStore)); 1.2739 + // Barriers are so large that short branch doesn't reach! 1.2740 + __ jmp(done); 1.2741 + 1.2742 + // Same code as above, but don't need rdx to test for volatile. 1.2743 + __ bind(notVolatile); 1.2744 + 1.2745 + if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx); 1.2746 + 1.2747 + // Get object from stack 1.2748 + pop_and_check_object(rcx); 1.2749 + 1.2750 + // access field 1.2751 + switch (bytecode()) { 1.2752 + case Bytecodes::_fast_bputfield: __ movb(lo, rax); break; 1.2753 + case Bytecodes::_fast_sputfield: // fall through 1.2754 + case Bytecodes::_fast_cputfield: __ movw(lo, rax); break; 1.2755 + case Bytecodes::_fast_iputfield: __ movl(lo, rax); break; 1.2756 + case Bytecodes::_fast_lputfield: 1.2757 + NOT_LP64(__ movptr(hi, rdx)); 1.2758 + __ movptr(lo, rax); 1.2759 + break; 1.2760 + case Bytecodes::_fast_fputfield: __ fstp_s(lo); break; 1.2761 + case Bytecodes::_fast_dputfield: __ fstp_d(lo); break; 1.2762 + case Bytecodes::_fast_aputfield: { 1.2763 + do_oop_store(_masm, lo, rax, _bs->kind(), false); 1.2764 + break; 1.2765 + } 1.2766 + default: 1.2767 + ShouldNotReachHere(); 1.2768 + } 1.2769 + __ bind(done); 1.2770 +} 1.2771 + 1.2772 + 1.2773 +void TemplateTable::fast_accessfield(TosState state) { 1.2774 + transition(atos, state); 1.2775 + 1.2776 + // do the JVMTI work here to avoid disturbing the register state below 1.2777 + if (JvmtiExport::can_post_field_access()) { 1.2778 + // Check to see if a field access watch has been set before we take 1.2779 + // the time to call into the VM. 1.2780 + Label L1; 1.2781 + __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr())); 1.2782 + __ testl(rcx,rcx); 1.2783 + __ jcc(Assembler::zero, L1); 1.2784 + // access constant pool cache entry 1.2785 + __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1); 1.2786 + __ push_ptr(rax); // save object pointer before call_VM() clobbers it 1.2787 + __ verify_oop(rax); 1.2788 + // rax,: object pointer copied above 1.2789 + // rcx: cache entry pointer 1.2790 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx); 1.2791 + __ pop_ptr(rax); // restore object pointer 1.2792 + __ bind(L1); 1.2793 + } 1.2794 + 1.2795 + // access constant pool cache 1.2796 + __ get_cache_and_index_at_bcp(rcx, rbx, 1); 1.2797 + // replace index with field offset from cache entry 1.2798 + __ movptr(rbx, Address(rcx, 1.2799 + rbx, 1.2800 + Address::times_ptr, 1.2801 + in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()))); 1.2802 + 1.2803 + 1.2804 + // rax,: object 1.2805 + __ verify_oop(rax); 1.2806 + __ null_check(rax); 1.2807 + // field addresses 1.2808 + const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize); 1.2809 + const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize); 1.2810 + 1.2811 + // access field 1.2812 + switch (bytecode()) { 1.2813 + case Bytecodes::_fast_bgetfield: __ movsbl(rax, lo ); break; 1.2814 + case Bytecodes::_fast_sgetfield: __ load_signed_short(rax, lo ); break; 1.2815 + case Bytecodes::_fast_cgetfield: __ load_unsigned_short(rax, lo ); break; 1.2816 + case Bytecodes::_fast_igetfield: __ movl(rax, lo); break; 1.2817 + case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten"); break; 1.2818 + case Bytecodes::_fast_fgetfield: __ fld_s(lo); break; 1.2819 + case Bytecodes::_fast_dgetfield: __ fld_d(lo); break; 1.2820 + case Bytecodes::_fast_agetfield: __ movptr(rax, lo); __ verify_oop(rax); break; 1.2821 + default: 1.2822 + ShouldNotReachHere(); 1.2823 + } 1.2824 + 1.2825 + // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO) 1.2826 + // volatile_barrier( ); 1.2827 +} 1.2828 + 1.2829 +void TemplateTable::fast_xaccess(TosState state) { 1.2830 + transition(vtos, state); 1.2831 + // get receiver 1.2832 + __ movptr(rax, aaddress(0)); 1.2833 + // access constant pool cache 1.2834 + __ get_cache_and_index_at_bcp(rcx, rdx, 2); 1.2835 + __ movptr(rbx, Address(rcx, 1.2836 + rdx, 1.2837 + Address::times_ptr, 1.2838 + in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()))); 1.2839 + // make sure exception is reported in correct bcp range (getfield is next instruction) 1.2840 + __ increment(rsi); 1.2841 + __ null_check(rax); 1.2842 + const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize); 1.2843 + if (state == itos) { 1.2844 + __ movl(rax, lo); 1.2845 + } else if (state == atos) { 1.2846 + __ movptr(rax, lo); 1.2847 + __ verify_oop(rax); 1.2848 + } else if (state == ftos) { 1.2849 + __ fld_s(lo); 1.2850 + } else { 1.2851 + ShouldNotReachHere(); 1.2852 + } 1.2853 + __ decrement(rsi); 1.2854 +} 1.2855 + 1.2856 + 1.2857 + 1.2858 +//---------------------------------------------------------------------------------------------------- 1.2859 +// Calls 1.2860 + 1.2861 +void TemplateTable::count_calls(Register method, Register temp) { 1.2862 + // implemented elsewhere 1.2863 + ShouldNotReachHere(); 1.2864 +} 1.2865 + 1.2866 + 1.2867 +void TemplateTable::prepare_invoke(int byte_no, 1.2868 + Register method, // linked method (or i-klass) 1.2869 + Register index, // itable index, MethodType, etc. 1.2870 + Register recv, // if caller wants to see it 1.2871 + Register flags // if caller wants to test it 1.2872 + ) { 1.2873 + // determine flags 1.2874 + const Bytecodes::Code code = bytecode(); 1.2875 + const bool is_invokeinterface = code == Bytecodes::_invokeinterface; 1.2876 + const bool is_invokedynamic = code == Bytecodes::_invokedynamic; 1.2877 + const bool is_invokehandle = code == Bytecodes::_invokehandle; 1.2878 + const bool is_invokevirtual = code == Bytecodes::_invokevirtual; 1.2879 + const bool is_invokespecial = code == Bytecodes::_invokespecial; 1.2880 + const bool load_receiver = (recv != noreg); 1.2881 + const bool save_flags = (flags != noreg); 1.2882 + assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), ""); 1.2883 + assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal"); 1.2884 + assert(flags == noreg || flags == rdx, ""); 1.2885 + assert(recv == noreg || recv == rcx, ""); 1.2886 + 1.2887 + // setup registers & access constant pool cache 1.2888 + if (recv == noreg) recv = rcx; 1.2889 + if (flags == noreg) flags = rdx; 1.2890 + assert_different_registers(method, index, recv, flags); 1.2891 + 1.2892 + // save 'interpreter return address' 1.2893 + __ save_bcp(); 1.2894 + 1.2895 + load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic); 1.2896 + 1.2897 + // maybe push appendix to arguments (just before return address) 1.2898 + if (is_invokedynamic || is_invokehandle) { 1.2899 + Label L_no_push; 1.2900 + __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift)); 1.2901 + __ jccb(Assembler::zero, L_no_push); 1.2902 + // Push the appendix as a trailing parameter. 1.2903 + // This must be done before we get the receiver, 1.2904 + // since the parameter_size includes it. 1.2905 + __ push(rbx); 1.2906 + __ mov(rbx, index); 1.2907 + assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0"); 1.2908 + __ load_resolved_reference_at_index(index, rbx); 1.2909 + __ pop(rbx); 1.2910 + __ push(index); // push appendix (MethodType, CallSite, etc.) 1.2911 + __ bind(L_no_push); 1.2912 + } 1.2913 + 1.2914 + // load receiver if needed (note: no return address pushed yet) 1.2915 + if (load_receiver) { 1.2916 + __ movl(recv, flags); 1.2917 + __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask); 1.2918 + const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address 1.2919 + const int receiver_is_at_end = -1; // back off one slot to get receiver 1.2920 + Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end); 1.2921 + __ movptr(recv, recv_addr); 1.2922 + __ verify_oop(recv); 1.2923 + } 1.2924 + 1.2925 + if (save_flags) { 1.2926 + __ mov(rsi, flags); 1.2927 + } 1.2928 + 1.2929 + // compute return type 1.2930 + __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift); 1.2931 + // Make sure we don't need to mask flags after the above shift 1.2932 + ConstantPoolCacheEntry::verify_tos_state_shift(); 1.2933 + // load return address 1.2934 + { 1.2935 + const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code); 1.2936 + ExternalAddress table(table_addr); 1.2937 + __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))); 1.2938 + } 1.2939 + 1.2940 + // push return address 1.2941 + __ push(flags); 1.2942 + 1.2943 + // Restore flags value from the constant pool cache, and restore rsi 1.2944 + // for later null checks. rsi is the bytecode pointer 1.2945 + if (save_flags) { 1.2946 + __ mov(flags, rsi); 1.2947 + __ restore_bcp(); 1.2948 + } 1.2949 +} 1.2950 + 1.2951 + 1.2952 +void TemplateTable::invokevirtual_helper(Register index, 1.2953 + Register recv, 1.2954 + Register flags) { 1.2955 + // Uses temporary registers rax, rdx 1.2956 + assert_different_registers(index, recv, rax, rdx); 1.2957 + assert(index == rbx, ""); 1.2958 + assert(recv == rcx, ""); 1.2959 + 1.2960 + // Test for an invoke of a final method 1.2961 + Label notFinal; 1.2962 + __ movl(rax, flags); 1.2963 + __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift)); 1.2964 + __ jcc(Assembler::zero, notFinal); 1.2965 + 1.2966 + const Register method = index; // method must be rbx 1.2967 + assert(method == rbx, 1.2968 + "Method* must be rbx for interpreter calling convention"); 1.2969 + 1.2970 + // do the call - the index is actually the method to call 1.2971 + // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method* 1.2972 + 1.2973 + // It's final, need a null check here! 1.2974 + __ null_check(recv); 1.2975 + 1.2976 + // profile this call 1.2977 + __ profile_final_call(rax); 1.2978 + __ profile_arguments_type(rax, method, rsi, true); 1.2979 + 1.2980 + __ jump_from_interpreted(method, rax); 1.2981 + 1.2982 + __ bind(notFinal); 1.2983 + 1.2984 + // get receiver klass 1.2985 + __ null_check(recv, oopDesc::klass_offset_in_bytes()); 1.2986 + __ load_klass(rax, recv); 1.2987 + 1.2988 + // profile this call 1.2989 + __ profile_virtual_call(rax, rdi, rdx); 1.2990 + 1.2991 + // get target Method* & entry point 1.2992 + __ lookup_virtual_method(rax, index, method); 1.2993 + __ profile_arguments_type(rdx, method, rsi, true); 1.2994 + __ jump_from_interpreted(method, rdx); 1.2995 +} 1.2996 + 1.2997 + 1.2998 +void TemplateTable::invokevirtual(int byte_no) { 1.2999 + transition(vtos, vtos); 1.3000 + assert(byte_no == f2_byte, "use this argument"); 1.3001 + prepare_invoke(byte_no, 1.3002 + rbx, // method or vtable index 1.3003 + noreg, // unused itable index 1.3004 + rcx, rdx); // recv, flags 1.3005 + 1.3006 + // rbx: index 1.3007 + // rcx: receiver 1.3008 + // rdx: flags 1.3009 + 1.3010 + invokevirtual_helper(rbx, rcx, rdx); 1.3011 +} 1.3012 + 1.3013 + 1.3014 +void TemplateTable::invokespecial(int byte_no) { 1.3015 + transition(vtos, vtos); 1.3016 + assert(byte_no == f1_byte, "use this argument"); 1.3017 + prepare_invoke(byte_no, rbx, noreg, // get f1 Method* 1.3018 + rcx); // get receiver also for null check 1.3019 + __ verify_oop(rcx); 1.3020 + __ null_check(rcx); 1.3021 + // do the call 1.3022 + __ profile_call(rax); 1.3023 + __ profile_arguments_type(rax, rbx, rsi, false); 1.3024 + __ jump_from_interpreted(rbx, rax); 1.3025 +} 1.3026 + 1.3027 + 1.3028 +void TemplateTable::invokestatic(int byte_no) { 1.3029 + transition(vtos, vtos); 1.3030 + assert(byte_no == f1_byte, "use this argument"); 1.3031 + prepare_invoke(byte_no, rbx); // get f1 Method* 1.3032 + // do the call 1.3033 + __ profile_call(rax); 1.3034 + __ profile_arguments_type(rax, rbx, rsi, false); 1.3035 + __ jump_from_interpreted(rbx, rax); 1.3036 +} 1.3037 + 1.3038 + 1.3039 +void TemplateTable::fast_invokevfinal(int byte_no) { 1.3040 + transition(vtos, vtos); 1.3041 + assert(byte_no == f2_byte, "use this argument"); 1.3042 + __ stop("fast_invokevfinal not used on x86"); 1.3043 +} 1.3044 + 1.3045 + 1.3046 +void TemplateTable::invokeinterface(int byte_no) { 1.3047 + transition(vtos, vtos); 1.3048 + assert(byte_no == f1_byte, "use this argument"); 1.3049 + prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index 1.3050 + rcx, rdx); // recv, flags 1.3051 + 1.3052 + // rax: interface klass (from f1) 1.3053 + // rbx: itable index (from f2) 1.3054 + // rcx: receiver 1.3055 + // rdx: flags 1.3056 + 1.3057 + // Special case of invokeinterface called for virtual method of 1.3058 + // java.lang.Object. See cpCacheOop.cpp for details. 1.3059 + // This code isn't produced by javac, but could be produced by 1.3060 + // another compliant java compiler. 1.3061 + Label notMethod; 1.3062 + __ movl(rdi, rdx); 1.3063 + __ andl(rdi, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift)); 1.3064 + __ jcc(Assembler::zero, notMethod); 1.3065 + 1.3066 + invokevirtual_helper(rbx, rcx, rdx); 1.3067 + __ bind(notMethod); 1.3068 + 1.3069 + // Get receiver klass into rdx - also a null check 1.3070 + __ restore_locals(); // restore rdi 1.3071 + __ null_check(rcx, oopDesc::klass_offset_in_bytes()); 1.3072 + __ load_klass(rdx, rcx); 1.3073 + 1.3074 + // profile this call 1.3075 + __ profile_virtual_call(rdx, rsi, rdi); 1.3076 + 1.3077 + Label no_such_interface, no_such_method; 1.3078 + 1.3079 + __ lookup_interface_method(// inputs: rec. class, interface, itable index 1.3080 + rdx, rax, rbx, 1.3081 + // outputs: method, scan temp. reg 1.3082 + rbx, rsi, 1.3083 + no_such_interface); 1.3084 + 1.3085 + // rbx: Method* to call 1.3086 + // rcx: receiver 1.3087 + // Check for abstract method error 1.3088 + // Note: This should be done more efficiently via a throw_abstract_method_error 1.3089 + // interpreter entry point and a conditional jump to it in case of a null 1.3090 + // method. 1.3091 + __ testptr(rbx, rbx); 1.3092 + __ jcc(Assembler::zero, no_such_method); 1.3093 + 1.3094 + __ profile_arguments_type(rdx, rbx, rsi, true); 1.3095 + 1.3096 + // do the call 1.3097 + // rcx: receiver 1.3098 + // rbx,: Method* 1.3099 + __ jump_from_interpreted(rbx, rdx); 1.3100 + __ should_not_reach_here(); 1.3101 + 1.3102 + // exception handling code follows... 1.3103 + // note: must restore interpreter registers to canonical 1.3104 + // state for exception handling to work correctly! 1.3105 + 1.3106 + __ bind(no_such_method); 1.3107 + // throw exception 1.3108 + __ pop(rbx); // pop return address (pushed by prepare_invoke) 1.3109 + __ restore_bcp(); // rsi must be correct for exception handler (was destroyed) 1.3110 + __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 1.3111 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError)); 1.3112 + // the call_VM checks for exception, so we should never return here. 1.3113 + __ should_not_reach_here(); 1.3114 + 1.3115 + __ bind(no_such_interface); 1.3116 + // throw exception 1.3117 + __ pop(rbx); // pop return address (pushed by prepare_invoke) 1.3118 + __ restore_bcp(); // rsi must be correct for exception handler (was destroyed) 1.3119 + __ restore_locals(); // make sure locals pointer is correct as well (was destroyed) 1.3120 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, 1.3121 + InterpreterRuntime::throw_IncompatibleClassChangeError)); 1.3122 + // the call_VM checks for exception, so we should never return here. 1.3123 + __ should_not_reach_here(); 1.3124 +} 1.3125 + 1.3126 +void TemplateTable::invokehandle(int byte_no) { 1.3127 + transition(vtos, vtos); 1.3128 + assert(byte_no == f1_byte, "use this argument"); 1.3129 + const Register rbx_method = rbx; 1.3130 + const Register rax_mtype = rax; 1.3131 + const Register rcx_recv = rcx; 1.3132 + const Register rdx_flags = rdx; 1.3133 + 1.3134 + if (!EnableInvokeDynamic) { 1.3135 + // rewriter does not generate this bytecode 1.3136 + __ should_not_reach_here(); 1.3137 + return; 1.3138 + } 1.3139 + 1.3140 + prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv); 1.3141 + __ verify_method_ptr(rbx_method); 1.3142 + __ verify_oop(rcx_recv); 1.3143 + __ null_check(rcx_recv); 1.3144 + 1.3145 + // rax: MethodType object (from cpool->resolved_references[f1], if necessary) 1.3146 + // rbx: MH.invokeExact_MT method (from f2) 1.3147 + 1.3148 + // Note: rax_mtype is already pushed (if necessary) by prepare_invoke 1.3149 + 1.3150 + // FIXME: profile the LambdaForm also 1.3151 + __ profile_final_call(rax); 1.3152 + __ profile_arguments_type(rdx, rbx_method, rsi, true); 1.3153 + 1.3154 + __ jump_from_interpreted(rbx_method, rdx); 1.3155 +} 1.3156 + 1.3157 + 1.3158 +void TemplateTable::invokedynamic(int byte_no) { 1.3159 + transition(vtos, vtos); 1.3160 + assert(byte_no == f1_byte, "use this argument"); 1.3161 + 1.3162 + if (!EnableInvokeDynamic) { 1.3163 + // We should not encounter this bytecode if !EnableInvokeDynamic. 1.3164 + // The verifier will stop it. However, if we get past the verifier, 1.3165 + // this will stop the thread in a reasonable way, without crashing the JVM. 1.3166 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, 1.3167 + InterpreterRuntime::throw_IncompatibleClassChangeError)); 1.3168 + // the call_VM checks for exception, so we should never return here. 1.3169 + __ should_not_reach_here(); 1.3170 + return; 1.3171 + } 1.3172 + 1.3173 + const Register rbx_method = rbx; 1.3174 + const Register rax_callsite = rax; 1.3175 + 1.3176 + prepare_invoke(byte_no, rbx_method, rax_callsite); 1.3177 + 1.3178 + // rax: CallSite object (from cpool->resolved_references[f1]) 1.3179 + // rbx: MH.linkToCallSite method (from f2) 1.3180 + 1.3181 + // Note: rax_callsite is already pushed by prepare_invoke 1.3182 + 1.3183 + // %%% should make a type profile for any invokedynamic that takes a ref argument 1.3184 + // profile this call 1.3185 + __ profile_call(rsi); 1.3186 + __ profile_arguments_type(rdx, rbx, rsi, false); 1.3187 + 1.3188 + __ verify_oop(rax_callsite); 1.3189 + 1.3190 + __ jump_from_interpreted(rbx_method, rdx); 1.3191 +} 1.3192 + 1.3193 +//---------------------------------------------------------------------------------------------------- 1.3194 +// Allocation 1.3195 + 1.3196 +void TemplateTable::_new() { 1.3197 + transition(vtos, atos); 1.3198 + __ get_unsigned_2_byte_index_at_bcp(rdx, 1); 1.3199 + Label slow_case; 1.3200 + Label slow_case_no_pop; 1.3201 + Label done; 1.3202 + Label initialize_header; 1.3203 + Label initialize_object; // including clearing the fields 1.3204 + Label allocate_shared; 1.3205 + 1.3206 + __ get_cpool_and_tags(rcx, rax); 1.3207 + 1.3208 + // Make sure the class we're about to instantiate has been resolved. 1.3209 + // This is done before loading InstanceKlass to be consistent with the order 1.3210 + // how Constant Pool is updated (see ConstantPool::klass_at_put) 1.3211 + const int tags_offset = Array<u1>::base_offset_in_bytes(); 1.3212 + __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class); 1.3213 + __ jcc(Assembler::notEqual, slow_case_no_pop); 1.3214 + 1.3215 + // get InstanceKlass 1.3216 + __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool))); 1.3217 + __ push(rcx); // save the contexts of klass for initializing the header 1.3218 + 1.3219 + // make sure klass is initialized & doesn't have finalizer 1.3220 + // make sure klass is fully initialized 1.3221 + __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized); 1.3222 + __ jcc(Assembler::notEqual, slow_case); 1.3223 + 1.3224 + // get instance_size in InstanceKlass (scaled to a count of bytes) 1.3225 + __ movl(rdx, Address(rcx, Klass::layout_helper_offset())); 1.3226 + // test to see if it has a finalizer or is malformed in some way 1.3227 + __ testl(rdx, Klass::_lh_instance_slow_path_bit); 1.3228 + __ jcc(Assembler::notZero, slow_case); 1.3229 + 1.3230 + // 1.3231 + // Allocate the instance 1.3232 + // 1) Try to allocate in the TLAB 1.3233 + // 2) if fail and the object is large allocate in the shared Eden 1.3234 + // 3) if the above fails (or is not applicable), go to a slow case 1.3235 + // (creates a new TLAB, etc.) 1.3236 + 1.3237 + const bool allow_shared_alloc = 1.3238 + Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode; 1.3239 + 1.3240 + const Register thread = rcx; 1.3241 + if (UseTLAB || allow_shared_alloc) { 1.3242 + __ get_thread(thread); 1.3243 + } 1.3244 + 1.3245 + if (UseTLAB) { 1.3246 + __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset()))); 1.3247 + __ lea(rbx, Address(rax, rdx, Address::times_1)); 1.3248 + __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset()))); 1.3249 + __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case); 1.3250 + __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx); 1.3251 + if (ZeroTLAB) { 1.3252 + // the fields have been already cleared 1.3253 + __ jmp(initialize_header); 1.3254 + } else { 1.3255 + // initialize both the header and fields 1.3256 + __ jmp(initialize_object); 1.3257 + } 1.3258 + } 1.3259 + 1.3260 + // Allocation in the shared Eden, if allowed. 1.3261 + // 1.3262 + // rdx: instance size in bytes 1.3263 + if (allow_shared_alloc) { 1.3264 + __ bind(allocate_shared); 1.3265 + 1.3266 + ExternalAddress heap_top((address)Universe::heap()->top_addr()); 1.3267 + 1.3268 + Label retry; 1.3269 + __ bind(retry); 1.3270 + __ movptr(rax, heap_top); 1.3271 + __ lea(rbx, Address(rax, rdx, Address::times_1)); 1.3272 + __ cmpptr(rbx, ExternalAddress((address)Universe::heap()->end_addr())); 1.3273 + __ jcc(Assembler::above, slow_case); 1.3274 + 1.3275 + // Compare rax, with the top addr, and if still equal, store the new 1.3276 + // top addr in rbx, at the address of the top addr pointer. Sets ZF if was 1.3277 + // equal, and clears it otherwise. Use lock prefix for atomicity on MPs. 1.3278 + // 1.3279 + // rax,: object begin 1.3280 + // rbx,: object end 1.3281 + // rdx: instance size in bytes 1.3282 + __ locked_cmpxchgptr(rbx, heap_top); 1.3283 + 1.3284 + // if someone beat us on the allocation, try again, otherwise continue 1.3285 + __ jcc(Assembler::notEqual, retry); 1.3286 + 1.3287 + __ incr_allocated_bytes(thread, rdx, 0); 1.3288 + } 1.3289 + 1.3290 + if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) { 1.3291 + // The object is initialized before the header. If the object size is 1.3292 + // zero, go directly to the header initialization. 1.3293 + __ bind(initialize_object); 1.3294 + __ decrement(rdx, sizeof(oopDesc)); 1.3295 + __ jcc(Assembler::zero, initialize_header); 1.3296 + 1.3297 + // Initialize topmost object field, divide rdx by 8, check if odd and 1.3298 + // test if zero. 1.3299 + __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code) 1.3300 + __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd 1.3301 + 1.3302 + // rdx must have been multiple of 8 1.3303 +#ifdef ASSERT 1.3304 + // make sure rdx was multiple of 8 1.3305 + Label L; 1.3306 + // Ignore partial flag stall after shrl() since it is debug VM 1.3307 + __ jccb(Assembler::carryClear, L); 1.3308 + __ stop("object size is not multiple of 2 - adjust this code"); 1.3309 + __ bind(L); 1.3310 + // rdx must be > 0, no extra check needed here 1.3311 +#endif 1.3312 + 1.3313 + // initialize remaining object fields: rdx was a multiple of 8 1.3314 + { Label loop; 1.3315 + __ bind(loop); 1.3316 + __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx); 1.3317 + NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx)); 1.3318 + __ decrement(rdx); 1.3319 + __ jcc(Assembler::notZero, loop); 1.3320 + } 1.3321 + 1.3322 + // initialize object header only. 1.3323 + __ bind(initialize_header); 1.3324 + if (UseBiasedLocking) { 1.3325 + __ pop(rcx); // get saved klass back in the register. 1.3326 + __ movptr(rbx, Address(rcx, Klass::prototype_header_offset())); 1.3327 + __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx); 1.3328 + } else { 1.3329 + __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), 1.3330 + (int32_t)markOopDesc::prototype()); // header 1.3331 + __ pop(rcx); // get saved klass back in the register. 1.3332 + } 1.3333 + __ store_klass(rax, rcx); // klass 1.3334 + 1.3335 + { 1.3336 + SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0); 1.3337 + // Trigger dtrace event for fastpath 1.3338 + __ push(atos); 1.3339 + __ call_VM_leaf( 1.3340 + CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax); 1.3341 + __ pop(atos); 1.3342 + } 1.3343 + 1.3344 + __ jmp(done); 1.3345 + } 1.3346 + 1.3347 + // slow case 1.3348 + __ bind(slow_case); 1.3349 + __ pop(rcx); // restore stack pointer to what it was when we came in. 1.3350 + __ bind(slow_case_no_pop); 1.3351 + __ get_constant_pool(rax); 1.3352 + __ get_unsigned_2_byte_index_at_bcp(rdx, 1); 1.3353 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx); 1.3354 + 1.3355 + // continue 1.3356 + __ bind(done); 1.3357 +} 1.3358 + 1.3359 + 1.3360 +void TemplateTable::newarray() { 1.3361 + transition(itos, atos); 1.3362 + __ push_i(rax); // make sure everything is on the stack 1.3363 + __ load_unsigned_byte(rdx, at_bcp(1)); 1.3364 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax); 1.3365 + __ pop_i(rdx); // discard size 1.3366 +} 1.3367 + 1.3368 + 1.3369 +void TemplateTable::anewarray() { 1.3370 + transition(itos, atos); 1.3371 + __ get_unsigned_2_byte_index_at_bcp(rdx, 1); 1.3372 + __ get_constant_pool(rcx); 1.3373 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax); 1.3374 +} 1.3375 + 1.3376 + 1.3377 +void TemplateTable::arraylength() { 1.3378 + transition(atos, itos); 1.3379 + __ null_check(rax, arrayOopDesc::length_offset_in_bytes()); 1.3380 + __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes())); 1.3381 +} 1.3382 + 1.3383 + 1.3384 +void TemplateTable::checkcast() { 1.3385 + transition(atos, atos); 1.3386 + Label done, is_null, ok_is_subtype, quicked, resolved; 1.3387 + __ testptr(rax, rax); // Object is in EAX 1.3388 + __ jcc(Assembler::zero, is_null); 1.3389 + 1.3390 + // Get cpool & tags index 1.3391 + __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array 1.3392 + __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index 1.3393 + // See if bytecode has already been quicked 1.3394 + __ cmpb(Address(rdx, rbx, Address::times_1, Array<u1>::base_offset_in_bytes()), JVM_CONSTANT_Class); 1.3395 + __ jcc(Assembler::equal, quicked); 1.3396 + 1.3397 + __ push(atos); 1.3398 + call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) ); 1.3399 + // vm_result_2 has metadata result 1.3400 + // borrow rdi from locals 1.3401 + __ get_thread(rdi); 1.3402 + __ get_vm_result_2(rax, rdi); 1.3403 + __ restore_locals(); 1.3404 + __ pop_ptr(rdx); 1.3405 + __ jmpb(resolved); 1.3406 + 1.3407 + // Get superklass in EAX and subklass in EBX 1.3408 + __ bind(quicked); 1.3409 + __ mov(rdx, rax); // Save object in EDX; EAX needed for subtype check 1.3410 + __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(ConstantPool))); 1.3411 + 1.3412 + __ bind(resolved); 1.3413 + __ load_klass(rbx, rdx); 1.3414 + 1.3415 + // Generate subtype check. Blows ECX. Resets EDI. Object in EDX. 1.3416 + // Superklass in EAX. Subklass in EBX. 1.3417 + __ gen_subtype_check( rbx, ok_is_subtype ); 1.3418 + 1.3419 + // Come here on failure 1.3420 + __ push(rdx); 1.3421 + // object is at TOS 1.3422 + __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry)); 1.3423 + 1.3424 + // Come here on success 1.3425 + __ bind(ok_is_subtype); 1.3426 + __ mov(rax,rdx); // Restore object in EDX 1.3427 + 1.3428 + // Collect counts on whether this check-cast sees NULLs a lot or not. 1.3429 + if (ProfileInterpreter) { 1.3430 + __ jmp(done); 1.3431 + __ bind(is_null); 1.3432 + __ profile_null_seen(rcx); 1.3433 + } else { 1.3434 + __ bind(is_null); // same as 'done' 1.3435 + } 1.3436 + __ bind(done); 1.3437 +} 1.3438 + 1.3439 + 1.3440 +void TemplateTable::instanceof() { 1.3441 + transition(atos, itos); 1.3442 + Label done, is_null, ok_is_subtype, quicked, resolved; 1.3443 + __ testptr(rax, rax); 1.3444 + __ jcc(Assembler::zero, is_null); 1.3445 + 1.3446 + // Get cpool & tags index 1.3447 + __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array 1.3448 + __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index 1.3449 + // See if bytecode has already been quicked 1.3450 + __ cmpb(Address(rdx, rbx, Address::times_1, Array<u1>::base_offset_in_bytes()), JVM_CONSTANT_Class); 1.3451 + __ jcc(Assembler::equal, quicked); 1.3452 + 1.3453 + __ push(atos); 1.3454 + call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) ); 1.3455 + // vm_result_2 has metadata result 1.3456 + // borrow rdi from locals 1.3457 + __ get_thread(rdi); 1.3458 + __ get_vm_result_2(rax, rdi); 1.3459 + __ restore_locals(); 1.3460 + __ pop_ptr(rdx); 1.3461 + __ load_klass(rdx, rdx); 1.3462 + __ jmp(resolved); 1.3463 + 1.3464 + // Get superklass in EAX and subklass in EDX 1.3465 + __ bind(quicked); 1.3466 + __ load_klass(rdx, rax); 1.3467 + __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(ConstantPool))); 1.3468 + 1.3469 + __ bind(resolved); 1.3470 + 1.3471 + // Generate subtype check. Blows ECX. Resets EDI. 1.3472 + // Superklass in EAX. Subklass in EDX. 1.3473 + __ gen_subtype_check( rdx, ok_is_subtype ); 1.3474 + 1.3475 + // Come here on failure 1.3476 + __ xorl(rax,rax); 1.3477 + __ jmpb(done); 1.3478 + // Come here on success 1.3479 + __ bind(ok_is_subtype); 1.3480 + __ movl(rax, 1); 1.3481 + 1.3482 + // Collect counts on whether this test sees NULLs a lot or not. 1.3483 + if (ProfileInterpreter) { 1.3484 + __ jmp(done); 1.3485 + __ bind(is_null); 1.3486 + __ profile_null_seen(rcx); 1.3487 + } else { 1.3488 + __ bind(is_null); // same as 'done' 1.3489 + } 1.3490 + __ bind(done); 1.3491 + // rax, = 0: obj == NULL or obj is not an instanceof the specified klass 1.3492 + // rax, = 1: obj != NULL and obj is an instanceof the specified klass 1.3493 +} 1.3494 + 1.3495 + 1.3496 +//---------------------------------------------------------------------------------------------------- 1.3497 +// Breakpoints 1.3498 +void TemplateTable::_breakpoint() { 1.3499 + 1.3500 + // Note: We get here even if we are single stepping.. 1.3501 + // jbug inists on setting breakpoints at every bytecode 1.3502 + // even if we are in single step mode. 1.3503 + 1.3504 + transition(vtos, vtos); 1.3505 + 1.3506 + // get the unpatched byte code 1.3507 + __ get_method(rcx); 1.3508 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi); 1.3509 + __ mov(rbx, rax); 1.3510 + 1.3511 + // post the breakpoint event 1.3512 + __ get_method(rcx); 1.3513 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi); 1.3514 + 1.3515 + // complete the execution of original bytecode 1.3516 + __ dispatch_only_normal(vtos); 1.3517 +} 1.3518 + 1.3519 + 1.3520 +//---------------------------------------------------------------------------------------------------- 1.3521 +// Exceptions 1.3522 + 1.3523 +void TemplateTable::athrow() { 1.3524 + transition(atos, vtos); 1.3525 + __ null_check(rax); 1.3526 + __ jump(ExternalAddress(Interpreter::throw_exception_entry())); 1.3527 +} 1.3528 + 1.3529 + 1.3530 +//---------------------------------------------------------------------------------------------------- 1.3531 +// Synchronization 1.3532 +// 1.3533 +// Note: monitorenter & exit are symmetric routines; which is reflected 1.3534 +// in the assembly code structure as well 1.3535 +// 1.3536 +// Stack layout: 1.3537 +// 1.3538 +// [expressions ] <--- rsp = expression stack top 1.3539 +// .. 1.3540 +// [expressions ] 1.3541 +// [monitor entry] <--- monitor block top = expression stack bot 1.3542 +// .. 1.3543 +// [monitor entry] 1.3544 +// [frame data ] <--- monitor block bot 1.3545 +// ... 1.3546 +// [saved rbp, ] <--- rbp, 1.3547 + 1.3548 + 1.3549 +void TemplateTable::monitorenter() { 1.3550 + transition(atos, vtos); 1.3551 + 1.3552 + // check for NULL object 1.3553 + __ null_check(rax); 1.3554 + 1.3555 + const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 1.3556 + const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 1.3557 + const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize); 1.3558 + Label allocated; 1.3559 + 1.3560 + // initialize entry pointer 1.3561 + __ xorl(rdx, rdx); // points to free slot or NULL 1.3562 + 1.3563 + // find a free slot in the monitor block (result in rdx) 1.3564 + { Label entry, loop, exit; 1.3565 + __ movptr(rcx, monitor_block_top); // points to current entry, starting with top-most entry 1.3566 + 1.3567 + __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block 1.3568 + __ jmpb(entry); 1.3569 + 1.3570 + __ bind(loop); 1.3571 + __ cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); // check if current entry is used 1.3572 + __ cmovptr(Assembler::equal, rdx, rcx); // if not used then remember entry in rdx 1.3573 + __ cmpptr(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object 1.3574 + __ jccb(Assembler::equal, exit); // if same object then stop searching 1.3575 + __ addptr(rcx, entry_size); // otherwise advance to next entry 1.3576 + __ bind(entry); 1.3577 + __ cmpptr(rcx, rbx); // check if bottom reached 1.3578 + __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry 1.3579 + __ bind(exit); 1.3580 + } 1.3581 + 1.3582 + __ testptr(rdx, rdx); // check if a slot has been found 1.3583 + __ jccb(Assembler::notZero, allocated); // if found, continue with that one 1.3584 + 1.3585 + // allocate one if there's no free slot 1.3586 + { Label entry, loop; 1.3587 + // 1. compute new pointers // rsp: old expression stack top 1.3588 + __ movptr(rdx, monitor_block_bot); // rdx: old expression stack bottom 1.3589 + __ subptr(rsp, entry_size); // move expression stack top 1.3590 + __ subptr(rdx, entry_size); // move expression stack bottom 1.3591 + __ mov(rcx, rsp); // set start value for copy loop 1.3592 + __ movptr(monitor_block_bot, rdx); // set new monitor block top 1.3593 + __ jmp(entry); 1.3594 + // 2. move expression stack contents 1.3595 + __ bind(loop); 1.3596 + __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location 1.3597 + __ movptr(Address(rcx, 0), rbx); // and store it at new location 1.3598 + __ addptr(rcx, wordSize); // advance to next word 1.3599 + __ bind(entry); 1.3600 + __ cmpptr(rcx, rdx); // check if bottom reached 1.3601 + __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word 1.3602 + } 1.3603 + 1.3604 + // call run-time routine 1.3605 + // rdx: points to monitor entry 1.3606 + __ bind(allocated); 1.3607 + 1.3608 + // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly. 1.3609 + // The object has already been poped from the stack, so the expression stack looks correct. 1.3610 + __ increment(rsi); 1.3611 + 1.3612 + __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax); // store object 1.3613 + __ lock_object(rdx); 1.3614 + 1.3615 + // check to make sure this monitor doesn't cause stack overflow after locking 1.3616 + __ save_bcp(); // in case of exception 1.3617 + __ generate_stack_overflow_check(0); 1.3618 + 1.3619 + // The bcp has already been incremented. Just need to dispatch to next instruction. 1.3620 + __ dispatch_next(vtos); 1.3621 +} 1.3622 + 1.3623 + 1.3624 +void TemplateTable::monitorexit() { 1.3625 + transition(atos, vtos); 1.3626 + 1.3627 + // check for NULL object 1.3628 + __ null_check(rax); 1.3629 + 1.3630 + const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 1.3631 + const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 1.3632 + const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize); 1.3633 + Label found; 1.3634 + 1.3635 + // find matching slot 1.3636 + { Label entry, loop; 1.3637 + __ movptr(rdx, monitor_block_top); // points to current entry, starting with top-most entry 1.3638 + __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block 1.3639 + __ jmpb(entry); 1.3640 + 1.3641 + __ bind(loop); 1.3642 + __ cmpptr(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object 1.3643 + __ jcc(Assembler::equal, found); // if same object then stop searching 1.3644 + __ addptr(rdx, entry_size); // otherwise advance to next entry 1.3645 + __ bind(entry); 1.3646 + __ cmpptr(rdx, rbx); // check if bottom reached 1.3647 + __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry 1.3648 + } 1.3649 + 1.3650 + // error handling. Unlocking was not block-structured 1.3651 + Label end; 1.3652 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); 1.3653 + __ should_not_reach_here(); 1.3654 + 1.3655 + // call run-time routine 1.3656 + // rcx: points to monitor entry 1.3657 + __ bind(found); 1.3658 + __ push_ptr(rax); // make sure object is on stack (contract with oopMaps) 1.3659 + __ unlock_object(rdx); 1.3660 + __ pop_ptr(rax); // discard object 1.3661 + __ bind(end); 1.3662 +} 1.3663 + 1.3664 + 1.3665 +//---------------------------------------------------------------------------------------------------- 1.3666 +// Wide instructions 1.3667 + 1.3668 +void TemplateTable::wide() { 1.3669 + transition(vtos, vtos); 1.3670 + __ load_unsigned_byte(rbx, at_bcp(1)); 1.3671 + ExternalAddress wtable((address)Interpreter::_wentry_point); 1.3672 + __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr))); 1.3673 + // Note: the rsi increment step is part of the individual wide bytecode implementations 1.3674 +} 1.3675 + 1.3676 + 1.3677 +//---------------------------------------------------------------------------------------------------- 1.3678 +// Multi arrays 1.3679 + 1.3680 +void TemplateTable::multianewarray() { 1.3681 + transition(vtos, atos); 1.3682 + __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions 1.3683 + // last dim is on top of stack; we want address of first one: 1.3684 + // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize 1.3685 + // the latter wordSize to point to the beginning of the array. 1.3686 + __ lea( rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize)); 1.3687 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax); // pass in rax, 1.3688 + __ load_unsigned_byte(rbx, at_bcp(3)); 1.3689 + __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts 1.3690 +} 1.3691 + 1.3692 +#endif /* !CC_INTERP */