1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/src/cpu/x86/vm/templateTable_x86_32.cpp Sat Dec 01 00:00:00 2007 +0000
1.3 @@ -0,0 +1,3499 @@
1.4 +/*
1.5 + * Copyright 1997-2007 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
1.23 + * CA 95054 USA or visit www.sun.com if you need additional information or
1.24 + * have any questions.
1.25 + *
1.26 + */
1.27 +
1.28 +#include "incls/_precompiled.incl"
1.29 +#include "incls/_templateTable_x86_32.cpp.incl"
1.30 +
1.31 +#ifndef CC_INTERP
1.32 +#define __ _masm->
1.33 +
1.34 +//----------------------------------------------------------------------------------------------------
1.35 +// Platform-dependent initialization
1.36 +
1.37 +void TemplateTable::pd_initialize() {
1.38 + // No i486 specific initialization
1.39 +}
1.40 +
1.41 +//----------------------------------------------------------------------------------------------------
1.42 +// Address computation
1.43 +
1.44 +// local variables
1.45 +static inline Address iaddress(int n) {
1.46 + return Address(rdi, Interpreter::local_offset_in_bytes(n));
1.47 +}
1.48 +
1.49 +static inline Address laddress(int n) { return iaddress(n + 1); }
1.50 +static inline Address haddress(int n) { return iaddress(n + 0); }
1.51 +static inline Address faddress(int n) { return iaddress(n); }
1.52 +static inline Address daddress(int n) { return laddress(n); }
1.53 +static inline Address aaddress(int n) { return iaddress(n); }
1.54 +
1.55 +static inline Address iaddress(Register r) {
1.56 + return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::value_offset_in_bytes());
1.57 +}
1.58 +static inline Address laddress(Register r) {
1.59 + return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(1));
1.60 +}
1.61 +static inline Address haddress(Register r) {
1.62 + return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
1.63 +}
1.64 +
1.65 +static inline Address faddress(Register r) { return iaddress(r); };
1.66 +static inline Address daddress(Register r) {
1.67 + assert(!TaggedStackInterpreter, "This doesn't work");
1.68 + return laddress(r);
1.69 +};
1.70 +static inline Address aaddress(Register r) { return iaddress(r); };
1.71 +
1.72 +// expression stack
1.73 +// (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
1.74 +// data beyond the rsp which is potentially unsafe in an MT environment;
1.75 +// an interrupt may overwrite that data.)
1.76 +static inline Address at_rsp () {
1.77 + return Address(rsp, 0);
1.78 +}
1.79 +
1.80 +// At top of Java expression stack which may be different than rsp(). It
1.81 +// isn't for category 1 objects.
1.82 +static inline Address at_tos () {
1.83 + Address tos = Address(rsp, Interpreter::expr_offset_in_bytes(0));
1.84 + return tos;
1.85 +}
1.86 +
1.87 +static inline Address at_tos_p1() {
1.88 + return Address(rsp, Interpreter::expr_offset_in_bytes(1));
1.89 +}
1.90 +
1.91 +static inline Address at_tos_p2() {
1.92 + return Address(rsp, Interpreter::expr_offset_in_bytes(2));
1.93 +}
1.94 +
1.95 +// Condition conversion
1.96 +static Assembler::Condition j_not(TemplateTable::Condition cc) {
1.97 + switch (cc) {
1.98 + case TemplateTable::equal : return Assembler::notEqual;
1.99 + case TemplateTable::not_equal : return Assembler::equal;
1.100 + case TemplateTable::less : return Assembler::greaterEqual;
1.101 + case TemplateTable::less_equal : return Assembler::greater;
1.102 + case TemplateTable::greater : return Assembler::lessEqual;
1.103 + case TemplateTable::greater_equal: return Assembler::less;
1.104 + }
1.105 + ShouldNotReachHere();
1.106 + return Assembler::zero;
1.107 +}
1.108 +
1.109 +
1.110 +//----------------------------------------------------------------------------------------------------
1.111 +// Miscelaneous helper routines
1.112 +
1.113 +Address TemplateTable::at_bcp(int offset) {
1.114 + assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
1.115 + return Address(rsi, offset);
1.116 +}
1.117 +
1.118 +
1.119 +void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc,
1.120 + Register scratch,
1.121 + bool load_bc_into_scratch/*=true*/) {
1.122 +
1.123 + if (!RewriteBytecodes) return;
1.124 + // the pair bytecodes have already done the load.
1.125 + if (load_bc_into_scratch) __ movl(bc, bytecode);
1.126 + Label patch_done;
1.127 + if (JvmtiExport::can_post_breakpoint()) {
1.128 + Label fast_patch;
1.129 + // if a breakpoint is present we can't rewrite the stream directly
1.130 + __ movzxb(scratch, at_bcp(0));
1.131 + __ cmpl(scratch, Bytecodes::_breakpoint);
1.132 + __ jcc(Assembler::notEqual, fast_patch);
1.133 + __ get_method(scratch);
1.134 + // Let breakpoint table handling rewrite to quicker bytecode
1.135 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, rsi, bc);
1.136 +#ifndef ASSERT
1.137 + __ jmpb(patch_done);
1.138 + __ bind(fast_patch);
1.139 + }
1.140 +#else
1.141 + __ jmp(patch_done);
1.142 + __ bind(fast_patch);
1.143 + }
1.144 + Label okay;
1.145 + __ load_unsigned_byte(scratch, at_bcp(0));
1.146 + __ cmpl(scratch, (int)Bytecodes::java_code(bytecode));
1.147 + __ jccb(Assembler::equal, okay);
1.148 + __ cmpl(scratch, bc);
1.149 + __ jcc(Assembler::equal, okay);
1.150 + __ stop("patching the wrong bytecode");
1.151 + __ bind(okay);
1.152 +#endif
1.153 + // patch bytecode
1.154 + __ movb(at_bcp(0), bc);
1.155 + __ bind(patch_done);
1.156 +}
1.157 +
1.158 +//----------------------------------------------------------------------------------------------------
1.159 +// Individual instructions
1.160 +
1.161 +void TemplateTable::nop() {
1.162 + transition(vtos, vtos);
1.163 + // nothing to do
1.164 +}
1.165 +
1.166 +void TemplateTable::shouldnotreachhere() {
1.167 + transition(vtos, vtos);
1.168 + __ stop("shouldnotreachhere bytecode");
1.169 +}
1.170 +
1.171 +
1.172 +
1.173 +void TemplateTable::aconst_null() {
1.174 + transition(vtos, atos);
1.175 + __ xorl(rax, rax);
1.176 +}
1.177 +
1.178 +
1.179 +void TemplateTable::iconst(int value) {
1.180 + transition(vtos, itos);
1.181 + if (value == 0) {
1.182 + __ xorl(rax, rax);
1.183 + } else {
1.184 + __ movl(rax, value);
1.185 + }
1.186 +}
1.187 +
1.188 +
1.189 +void TemplateTable::lconst(int value) {
1.190 + transition(vtos, ltos);
1.191 + if (value == 0) {
1.192 + __ xorl(rax, rax);
1.193 + } else {
1.194 + __ movl(rax, value);
1.195 + }
1.196 + assert(value >= 0, "check this code");
1.197 + __ xorl(rdx, rdx);
1.198 +}
1.199 +
1.200 +
1.201 +void TemplateTable::fconst(int value) {
1.202 + transition(vtos, ftos);
1.203 + if (value == 0) { __ fldz();
1.204 + } else if (value == 1) { __ fld1();
1.205 + } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
1.206 + } else { ShouldNotReachHere();
1.207 + }
1.208 +}
1.209 +
1.210 +
1.211 +void TemplateTable::dconst(int value) {
1.212 + transition(vtos, dtos);
1.213 + if (value == 0) { __ fldz();
1.214 + } else if (value == 1) { __ fld1();
1.215 + } else { ShouldNotReachHere();
1.216 + }
1.217 +}
1.218 +
1.219 +
1.220 +void TemplateTable::bipush() {
1.221 + transition(vtos, itos);
1.222 + __ load_signed_byte(rax, at_bcp(1));
1.223 +}
1.224 +
1.225 +
1.226 +void TemplateTable::sipush() {
1.227 + transition(vtos, itos);
1.228 + __ load_unsigned_word(rax, at_bcp(1));
1.229 + __ bswap(rax);
1.230 + __ sarl(rax, 16);
1.231 +}
1.232 +
1.233 +void TemplateTable::ldc(bool wide) {
1.234 + transition(vtos, vtos);
1.235 + Label call_ldc, notFloat, notClass, Done;
1.236 +
1.237 + if (wide) {
1.238 + __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
1.239 + } else {
1.240 + __ load_unsigned_byte(rbx, at_bcp(1));
1.241 + }
1.242 + __ get_cpool_and_tags(rcx, rax);
1.243 + const int base_offset = constantPoolOopDesc::header_size() * wordSize;
1.244 + const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
1.245 +
1.246 + // get type
1.247 + __ xorl(rdx, rdx);
1.248 + __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset));
1.249 +
1.250 + // unresolved string - get the resolved string
1.251 + __ cmpl(rdx, JVM_CONSTANT_UnresolvedString);
1.252 + __ jccb(Assembler::equal, call_ldc);
1.253 +
1.254 + // unresolved class - get the resolved class
1.255 + __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
1.256 + __ jccb(Assembler::equal, call_ldc);
1.257 +
1.258 + // unresolved class in error (resolution failed) - call into runtime
1.259 + // so that the same error from first resolution attempt is thrown.
1.260 + __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
1.261 + __ jccb(Assembler::equal, call_ldc);
1.262 +
1.263 + // resolved class - need to call vm to get java mirror of the class
1.264 + __ cmpl(rdx, JVM_CONSTANT_Class);
1.265 + __ jcc(Assembler::notEqual, notClass);
1.266 +
1.267 + __ bind(call_ldc);
1.268 + __ movl(rcx, wide);
1.269 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx);
1.270 + __ push(atos);
1.271 + __ jmp(Done);
1.272 +
1.273 + __ bind(notClass);
1.274 + __ cmpl(rdx, JVM_CONSTANT_Float);
1.275 + __ jccb(Assembler::notEqual, notFloat);
1.276 + // ftos
1.277 + __ fld_s( Address(rcx, rbx, Address::times_4, base_offset));
1.278 + __ push(ftos);
1.279 + __ jmp(Done);
1.280 +
1.281 + __ bind(notFloat);
1.282 +#ifdef ASSERT
1.283 + { Label L;
1.284 + __ cmpl(rdx, JVM_CONSTANT_Integer);
1.285 + __ jcc(Assembler::equal, L);
1.286 + __ cmpl(rdx, JVM_CONSTANT_String);
1.287 + __ jcc(Assembler::equal, L);
1.288 + __ stop("unexpected tag type in ldc");
1.289 + __ bind(L);
1.290 + }
1.291 +#endif
1.292 + Label isOop;
1.293 + // atos and itos
1.294 + __ movl(rax, Address(rcx, rbx, Address::times_4, base_offset));
1.295 + // String is only oop type we will see here
1.296 + __ cmpl(rdx, JVM_CONSTANT_String);
1.297 + __ jccb(Assembler::equal, isOop);
1.298 + __ push(itos);
1.299 + __ jmp(Done);
1.300 + __ bind(isOop);
1.301 + __ push(atos);
1.302 +
1.303 + if (VerifyOops) {
1.304 + __ verify_oop(rax);
1.305 + }
1.306 + __ bind(Done);
1.307 +}
1.308 +
1.309 +void TemplateTable::ldc2_w() {
1.310 + transition(vtos, vtos);
1.311 + Label Long, Done;
1.312 + __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
1.313 +
1.314 + __ get_cpool_and_tags(rcx, rax);
1.315 + const int base_offset = constantPoolOopDesc::header_size() * wordSize;
1.316 + const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
1.317 +
1.318 + // get type
1.319 + __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double);
1.320 + __ jccb(Assembler::notEqual, Long);
1.321 + // dtos
1.322 + __ fld_d( Address(rcx, rbx, Address::times_4, base_offset));
1.323 + __ push(dtos);
1.324 + __ jmpb(Done);
1.325 +
1.326 + __ bind(Long);
1.327 + // ltos
1.328 + __ movl(rax, Address(rcx, rbx, Address::times_4, base_offset + 0 * wordSize));
1.329 + __ movl(rdx, Address(rcx, rbx, Address::times_4, base_offset + 1 * wordSize));
1.330 +
1.331 + __ push(ltos);
1.332 +
1.333 + __ bind(Done);
1.334 +}
1.335 +
1.336 +
1.337 +void TemplateTable::locals_index(Register reg, int offset) {
1.338 + __ load_unsigned_byte(reg, at_bcp(offset));
1.339 + __ negl(reg);
1.340 +}
1.341 +
1.342 +
1.343 +void TemplateTable::iload() {
1.344 + transition(vtos, itos);
1.345 + if (RewriteFrequentPairs) {
1.346 + Label rewrite, done;
1.347 +
1.348 + // get next byte
1.349 + __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
1.350 + // if _iload, wait to rewrite to iload2. We only want to rewrite the
1.351 + // last two iloads in a pair. Comparing against fast_iload means that
1.352 + // the next bytecode is neither an iload or a caload, and therefore
1.353 + // an iload pair.
1.354 + __ cmpl(rbx, Bytecodes::_iload);
1.355 + __ jcc(Assembler::equal, done);
1.356 +
1.357 + __ cmpl(rbx, Bytecodes::_fast_iload);
1.358 + __ movl(rcx, Bytecodes::_fast_iload2);
1.359 + __ jccb(Assembler::equal, rewrite);
1.360 +
1.361 + // if _caload, rewrite to fast_icaload
1.362 + __ cmpl(rbx, Bytecodes::_caload);
1.363 + __ movl(rcx, Bytecodes::_fast_icaload);
1.364 + __ jccb(Assembler::equal, rewrite);
1.365 +
1.366 + // rewrite so iload doesn't check again.
1.367 + __ movl(rcx, Bytecodes::_fast_iload);
1.368 +
1.369 + // rewrite
1.370 + // rcx: fast bytecode
1.371 + __ bind(rewrite);
1.372 + patch_bytecode(Bytecodes::_iload, rcx, rbx, false);
1.373 + __ bind(done);
1.374 + }
1.375 +
1.376 + // Get the local value into tos
1.377 + locals_index(rbx);
1.378 + __ movl(rax, iaddress(rbx));
1.379 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.380 +}
1.381 +
1.382 +
1.383 +void TemplateTable::fast_iload2() {
1.384 + transition(vtos, itos);
1.385 + locals_index(rbx);
1.386 + __ movl(rax, iaddress(rbx));
1.387 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.388 + __ push(itos);
1.389 + locals_index(rbx, 3);
1.390 + __ movl(rax, iaddress(rbx));
1.391 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.392 +}
1.393 +
1.394 +void TemplateTable::fast_iload() {
1.395 + transition(vtos, itos);
1.396 + locals_index(rbx);
1.397 + __ movl(rax, iaddress(rbx));
1.398 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.399 +}
1.400 +
1.401 +
1.402 +void TemplateTable::lload() {
1.403 + transition(vtos, ltos);
1.404 + locals_index(rbx);
1.405 + __ movl(rax, laddress(rbx));
1.406 + __ movl(rdx, haddress(rbx));
1.407 + debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
1.408 +}
1.409 +
1.410 +
1.411 +void TemplateTable::fload() {
1.412 + transition(vtos, ftos);
1.413 + locals_index(rbx);
1.414 + __ fld_s(faddress(rbx));
1.415 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.416 +}
1.417 +
1.418 +
1.419 +void TemplateTable::dload() {
1.420 + transition(vtos, dtos);
1.421 + locals_index(rbx);
1.422 + if (TaggedStackInterpreter) {
1.423 + // Get double out of locals array, onto temp stack and load with
1.424 + // float instruction into ST0
1.425 + __ movl(rax, laddress(rbx));
1.426 + __ movl(rdx, haddress(rbx));
1.427 + __ pushl(rdx); // push hi first
1.428 + __ pushl(rax);
1.429 + __ fld_d(Address(rsp, 0));
1.430 + __ addl(rsp, 2*wordSize);
1.431 + debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
1.432 + } else {
1.433 + __ fld_d(daddress(rbx));
1.434 + }
1.435 +}
1.436 +
1.437 +
1.438 +void TemplateTable::aload() {
1.439 + transition(vtos, atos);
1.440 + locals_index(rbx);
1.441 + __ movl(rax, iaddress(rbx));
1.442 + debug_only(__ verify_local_tag(frame::TagReference, rbx));
1.443 +}
1.444 +
1.445 +
1.446 +void TemplateTable::locals_index_wide(Register reg) {
1.447 + __ movl(reg, at_bcp(2));
1.448 + __ bswap(reg);
1.449 + __ shrl(reg, 16);
1.450 + __ negl(reg);
1.451 +}
1.452 +
1.453 +
1.454 +void TemplateTable::wide_iload() {
1.455 + transition(vtos, itos);
1.456 + locals_index_wide(rbx);
1.457 + __ movl(rax, iaddress(rbx));
1.458 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.459 +}
1.460 +
1.461 +
1.462 +void TemplateTable::wide_lload() {
1.463 + transition(vtos, ltos);
1.464 + locals_index_wide(rbx);
1.465 + __ movl(rax, laddress(rbx));
1.466 + __ movl(rdx, haddress(rbx));
1.467 + debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
1.468 +}
1.469 +
1.470 +
1.471 +void TemplateTable::wide_fload() {
1.472 + transition(vtos, ftos);
1.473 + locals_index_wide(rbx);
1.474 + __ fld_s(faddress(rbx));
1.475 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.476 +}
1.477 +
1.478 +
1.479 +void TemplateTable::wide_dload() {
1.480 + transition(vtos, dtos);
1.481 + locals_index_wide(rbx);
1.482 + if (TaggedStackInterpreter) {
1.483 + // Get double out of locals array, onto temp stack and load with
1.484 + // float instruction into ST0
1.485 + __ movl(rax, laddress(rbx));
1.486 + __ movl(rdx, haddress(rbx));
1.487 + __ pushl(rdx); // push hi first
1.488 + __ pushl(rax);
1.489 + __ fld_d(Address(rsp, 0));
1.490 + __ addl(rsp, 2*wordSize);
1.491 + debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
1.492 + } else {
1.493 + __ fld_d(daddress(rbx));
1.494 + }
1.495 +}
1.496 +
1.497 +
1.498 +void TemplateTable::wide_aload() {
1.499 + transition(vtos, atos);
1.500 + locals_index_wide(rbx);
1.501 + __ movl(rax, iaddress(rbx));
1.502 + debug_only(__ verify_local_tag(frame::TagReference, rbx));
1.503 +}
1.504 +
1.505 +void TemplateTable::index_check(Register array, Register index) {
1.506 + // Pop ptr into array
1.507 + __ pop_ptr(array);
1.508 + index_check_without_pop(array, index);
1.509 +}
1.510 +
1.511 +void TemplateTable::index_check_without_pop(Register array, Register index) {
1.512 + // destroys rbx,
1.513 + // check array
1.514 + __ null_check(array, arrayOopDesc::length_offset_in_bytes());
1.515 + // check index
1.516 + __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
1.517 + if (index != rbx) {
1.518 + // ??? convention: move aberrant index into rbx, for exception message
1.519 + assert(rbx != array, "different registers");
1.520 + __ movl(rbx, index);
1.521 + }
1.522 + __ jump_cc(Assembler::aboveEqual,
1.523 + ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
1.524 +}
1.525 +
1.526 +
1.527 +void TemplateTable::iaload() {
1.528 + transition(itos, itos);
1.529 + // rdx: array
1.530 + index_check(rdx, rax); // kills rbx,
1.531 + // rax,: index
1.532 + __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)));
1.533 +}
1.534 +
1.535 +
1.536 +void TemplateTable::laload() {
1.537 + transition(itos, ltos);
1.538 + // rax,: index
1.539 + // rdx: array
1.540 + index_check(rdx, rax);
1.541 + __ movl(rbx, rax);
1.542 + // rbx,: index
1.543 + __ movl(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
1.544 + __ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize));
1.545 +}
1.546 +
1.547 +
1.548 +void TemplateTable::faload() {
1.549 + transition(itos, ftos);
1.550 + // rdx: array
1.551 + index_check(rdx, rax); // kills rbx,
1.552 + // rax,: index
1.553 + __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1.554 +}
1.555 +
1.556 +
1.557 +void TemplateTable::daload() {
1.558 + transition(itos, dtos);
1.559 + // rdx: array
1.560 + index_check(rdx, rax); // kills rbx,
1.561 + // rax,: index
1.562 + __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1.563 +}
1.564 +
1.565 +
1.566 +void TemplateTable::aaload() {
1.567 + transition(itos, atos);
1.568 + // rdx: array
1.569 + index_check(rdx, rax); // kills rbx,
1.570 + // rax,: index
1.571 + __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1.572 +}
1.573 +
1.574 +
1.575 +void TemplateTable::baload() {
1.576 + transition(itos, itos);
1.577 + // rdx: array
1.578 + index_check(rdx, rax); // kills rbx,
1.579 + // rax,: index
1.580 + // can do better code for P5 - fix this at some point
1.581 + __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
1.582 + __ movl(rax, rbx);
1.583 +}
1.584 +
1.585 +
1.586 +void TemplateTable::caload() {
1.587 + transition(itos, itos);
1.588 + // rdx: array
1.589 + index_check(rdx, rax); // kills rbx,
1.590 + // rax,: index
1.591 + // can do better code for P5 - may want to improve this at some point
1.592 + __ load_unsigned_word(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1.593 + __ movl(rax, rbx);
1.594 +}
1.595 +
1.596 +// iload followed by caload frequent pair
1.597 +void TemplateTable::fast_icaload() {
1.598 + transition(vtos, itos);
1.599 + // load index out of locals
1.600 + locals_index(rbx);
1.601 + __ movl(rax, iaddress(rbx));
1.602 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.603 +
1.604 + // rdx: array
1.605 + index_check(rdx, rax);
1.606 + // rax,: index
1.607 + __ load_unsigned_word(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1.608 + __ movl(rax, rbx);
1.609 +}
1.610 +
1.611 +void TemplateTable::saload() {
1.612 + transition(itos, itos);
1.613 + // rdx: array
1.614 + index_check(rdx, rax); // kills rbx,
1.615 + // rax,: index
1.616 + // can do better code for P5 - may want to improve this at some point
1.617 + __ load_signed_word(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
1.618 + __ movl(rax, rbx);
1.619 +}
1.620 +
1.621 +
1.622 +void TemplateTable::iload(int n) {
1.623 + transition(vtos, itos);
1.624 + __ movl(rax, iaddress(n));
1.625 + debug_only(__ verify_local_tag(frame::TagValue, n));
1.626 +}
1.627 +
1.628 +
1.629 +void TemplateTable::lload(int n) {
1.630 + transition(vtos, ltos);
1.631 + __ movl(rax, laddress(n));
1.632 + __ movl(rdx, haddress(n));
1.633 + debug_only(__ verify_local_tag(frame::TagCategory2, n));
1.634 +}
1.635 +
1.636 +
1.637 +void TemplateTable::fload(int n) {
1.638 + transition(vtos, ftos);
1.639 + __ fld_s(faddress(n));
1.640 + debug_only(__ verify_local_tag(frame::TagValue, n));
1.641 +}
1.642 +
1.643 +
1.644 +void TemplateTable::dload(int n) {
1.645 + transition(vtos, dtos);
1.646 + if (TaggedStackInterpreter) {
1.647 + // Get double out of locals array, onto temp stack and load with
1.648 + // float instruction into ST0
1.649 + __ movl(rax, laddress(n));
1.650 + __ movl(rdx, haddress(n));
1.651 + __ pushl(rdx); // push hi first
1.652 + __ pushl(rax);
1.653 + __ fld_d(Address(rsp, 0));
1.654 + __ addl(rsp, 2*wordSize); // reset rsp
1.655 + debug_only(__ verify_local_tag(frame::TagCategory2, n));
1.656 + } else {
1.657 + __ fld_d(daddress(n));
1.658 + }
1.659 +}
1.660 +
1.661 +
1.662 +void TemplateTable::aload(int n) {
1.663 + transition(vtos, atos);
1.664 + __ movl(rax, aaddress(n));
1.665 + debug_only(__ verify_local_tag(frame::TagReference, n));
1.666 +}
1.667 +
1.668 +
1.669 +void TemplateTable::aload_0() {
1.670 + transition(vtos, atos);
1.671 + // According to bytecode histograms, the pairs:
1.672 + //
1.673 + // _aload_0, _fast_igetfield
1.674 + // _aload_0, _fast_agetfield
1.675 + // _aload_0, _fast_fgetfield
1.676 + //
1.677 + // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
1.678 + // bytecode checks if the next bytecode is either _fast_igetfield,
1.679 + // _fast_agetfield or _fast_fgetfield and then rewrites the
1.680 + // current bytecode into a pair bytecode; otherwise it rewrites the current
1.681 + // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
1.682 + //
1.683 + // Note: If the next bytecode is _getfield, the rewrite must be delayed,
1.684 + // otherwise we may miss an opportunity for a pair.
1.685 + //
1.686 + // Also rewrite frequent pairs
1.687 + // aload_0, aload_1
1.688 + // aload_0, iload_1
1.689 + // These bytecodes with a small amount of code are most profitable to rewrite
1.690 + if (RewriteFrequentPairs) {
1.691 + Label rewrite, done;
1.692 + // get next byte
1.693 + __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
1.694 +
1.695 + // do actual aload_0
1.696 + aload(0);
1.697 +
1.698 + // if _getfield then wait with rewrite
1.699 + __ cmpl(rbx, Bytecodes::_getfield);
1.700 + __ jcc(Assembler::equal, done);
1.701 +
1.702 + // if _igetfield then reqrite to _fast_iaccess_0
1.703 + assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
1.704 + __ cmpl(rbx, Bytecodes::_fast_igetfield);
1.705 + __ movl(rcx, Bytecodes::_fast_iaccess_0);
1.706 + __ jccb(Assembler::equal, rewrite);
1.707 +
1.708 + // if _agetfield then reqrite to _fast_aaccess_0
1.709 + assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
1.710 + __ cmpl(rbx, Bytecodes::_fast_agetfield);
1.711 + __ movl(rcx, Bytecodes::_fast_aaccess_0);
1.712 + __ jccb(Assembler::equal, rewrite);
1.713 +
1.714 + // if _fgetfield then reqrite to _fast_faccess_0
1.715 + assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
1.716 + __ cmpl(rbx, Bytecodes::_fast_fgetfield);
1.717 + __ movl(rcx, Bytecodes::_fast_faccess_0);
1.718 + __ jccb(Assembler::equal, rewrite);
1.719 +
1.720 + // else rewrite to _fast_aload0
1.721 + assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
1.722 + __ movl(rcx, Bytecodes::_fast_aload_0);
1.723 +
1.724 + // rewrite
1.725 + // rcx: fast bytecode
1.726 + __ bind(rewrite);
1.727 + patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false);
1.728 +
1.729 + __ bind(done);
1.730 + } else {
1.731 + aload(0);
1.732 + }
1.733 +}
1.734 +
1.735 +void TemplateTable::istore() {
1.736 + transition(itos, vtos);
1.737 + locals_index(rbx);
1.738 + __ movl(iaddress(rbx), rax);
1.739 + __ tag_local(frame::TagValue, rbx);
1.740 +}
1.741 +
1.742 +
1.743 +void TemplateTable::lstore() {
1.744 + transition(ltos, vtos);
1.745 + locals_index(rbx);
1.746 + __ movl(laddress(rbx), rax);
1.747 + __ movl(haddress(rbx), rdx);
1.748 + __ tag_local(frame::TagCategory2, rbx);
1.749 +}
1.750 +
1.751 +
1.752 +void TemplateTable::fstore() {
1.753 + transition(ftos, vtos);
1.754 + locals_index(rbx);
1.755 + __ fstp_s(faddress(rbx));
1.756 + __ tag_local(frame::TagValue, rbx);
1.757 +}
1.758 +
1.759 +
1.760 +void TemplateTable::dstore() {
1.761 + transition(dtos, vtos);
1.762 + locals_index(rbx);
1.763 + if (TaggedStackInterpreter) {
1.764 + // Store double on stack and reload into locals nonadjacently
1.765 + __ subl(rsp, 2 * wordSize);
1.766 + __ fstp_d(Address(rsp, 0));
1.767 + __ popl(rax);
1.768 + __ popl(rdx);
1.769 + __ movl(laddress(rbx), rax);
1.770 + __ movl(haddress(rbx), rdx);
1.771 + __ tag_local(frame::TagCategory2, rbx);
1.772 + } else {
1.773 + __ fstp_d(daddress(rbx));
1.774 + }
1.775 +}
1.776 +
1.777 +
1.778 +void TemplateTable::astore() {
1.779 + transition(vtos, vtos);
1.780 + __ pop_ptr(rax, rdx); // will need to pop tag too
1.781 + locals_index(rbx);
1.782 + __ movl(aaddress(rbx), rax);
1.783 + __ tag_local(rdx, rbx); // need to store same tag in local may be returnAddr
1.784 +}
1.785 +
1.786 +
1.787 +void TemplateTable::wide_istore() {
1.788 + transition(vtos, vtos);
1.789 + __ pop_i(rax);
1.790 + locals_index_wide(rbx);
1.791 + __ movl(iaddress(rbx), rax);
1.792 + __ tag_local(frame::TagValue, rbx);
1.793 +}
1.794 +
1.795 +
1.796 +void TemplateTable::wide_lstore() {
1.797 + transition(vtos, vtos);
1.798 + __ pop_l(rax, rdx);
1.799 + locals_index_wide(rbx);
1.800 + __ movl(laddress(rbx), rax);
1.801 + __ movl(haddress(rbx), rdx);
1.802 + __ tag_local(frame::TagCategory2, rbx);
1.803 +}
1.804 +
1.805 +
1.806 +void TemplateTable::wide_fstore() {
1.807 + wide_istore();
1.808 +}
1.809 +
1.810 +
1.811 +void TemplateTable::wide_dstore() {
1.812 + wide_lstore();
1.813 +}
1.814 +
1.815 +
1.816 +void TemplateTable::wide_astore() {
1.817 + transition(vtos, vtos);
1.818 + __ pop_ptr(rax, rdx);
1.819 + locals_index_wide(rbx);
1.820 + __ movl(aaddress(rbx), rax);
1.821 + __ tag_local(rdx, rbx);
1.822 +}
1.823 +
1.824 +
1.825 +void TemplateTable::iastore() {
1.826 + transition(itos, vtos);
1.827 + __ pop_i(rbx);
1.828 + // rax,: value
1.829 + // rdx: array
1.830 + index_check(rdx, rbx); // prefer index in rbx,
1.831 + // rbx,: index
1.832 + __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax);
1.833 +}
1.834 +
1.835 +
1.836 +void TemplateTable::lastore() {
1.837 + transition(ltos, vtos);
1.838 + __ pop_i(rbx);
1.839 + // rax,: low(value)
1.840 + // rcx: array
1.841 + // rdx: high(value)
1.842 + index_check(rcx, rbx); // prefer index in rbx,
1.843 + // rbx,: index
1.844 + __ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1.845 + __ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx);
1.846 +}
1.847 +
1.848 +
1.849 +void TemplateTable::fastore() {
1.850 + transition(ftos, vtos);
1.851 + __ pop_i(rbx);
1.852 + // rdx: array
1.853 + // st0: value
1.854 + index_check(rdx, rbx); // prefer index in rbx,
1.855 + // rbx,: index
1.856 + __ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1.857 +}
1.858 +
1.859 +
1.860 +void TemplateTable::dastore() {
1.861 + transition(dtos, vtos);
1.862 + __ pop_i(rbx);
1.863 + // rdx: array
1.864 + // st0: value
1.865 + index_check(rdx, rbx); // prefer index in rbx,
1.866 + // rbx,: index
1.867 + __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1.868 +}
1.869 +
1.870 +
1.871 +void TemplateTable::aastore() {
1.872 + Label is_null, ok_is_subtype, done;
1.873 + transition(vtos, vtos);
1.874 + // stack: ..., array, index, value
1.875 + __ movl(rax, at_tos()); // Value
1.876 + __ movl(rcx, at_tos_p1()); // Index
1.877 + __ movl(rdx, at_tos_p2()); // Array
1.878 + index_check_without_pop(rdx, rcx); // kills rbx,
1.879 + // do array store check - check for NULL value first
1.880 + __ testl(rax, rax);
1.881 + __ jcc(Assembler::zero, is_null);
1.882 +
1.883 + // Move subklass into EBX
1.884 + __ movl(rbx, Address(rax, oopDesc::klass_offset_in_bytes()));
1.885 + // Move superklass into EAX
1.886 + __ movl(rax, Address(rdx, oopDesc::klass_offset_in_bytes()));
1.887 + __ movl(rax, Address(rax, sizeof(oopDesc) + objArrayKlass::element_klass_offset_in_bytes()));
1.888 + // Compress array+index*4+12 into a single register. Frees ECX.
1.889 + __ leal(rdx, Address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1.890 +
1.891 + // Generate subtype check. Blows ECX. Resets EDI to locals.
1.892 + // Superklass in EAX. Subklass in EBX.
1.893 + __ gen_subtype_check( rbx, ok_is_subtype );
1.894 +
1.895 + // Come here on failure
1.896 + // object is at TOS
1.897 + __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1.898 +
1.899 + // Come here on success
1.900 + __ bind(ok_is_subtype);
1.901 + __ movl(rax, at_rsp()); // Value
1.902 + __ movl(Address(rdx, 0), rax);
1.903 + __ store_check(rdx);
1.904 + __ jmpb(done);
1.905 +
1.906 + // Have a NULL in EAX, EDX=array, ECX=index. Store NULL at ary[idx]
1.907 + __ bind(is_null);
1.908 + __ profile_null_seen(rbx);
1.909 + __ movl(Address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT)), rax);
1.910 +
1.911 + // Pop stack arguments
1.912 + __ bind(done);
1.913 + __ addl(rsp, 3 * Interpreter::stackElementSize());
1.914 +}
1.915 +
1.916 +
1.917 +void TemplateTable::bastore() {
1.918 + transition(itos, vtos);
1.919 + __ pop_i(rbx);
1.920 + // rax,: value
1.921 + // rdx: array
1.922 + index_check(rdx, rbx); // prefer index in rbx,
1.923 + // rbx,: index
1.924 + __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax);
1.925 +}
1.926 +
1.927 +
1.928 +void TemplateTable::castore() {
1.929 + transition(itos, vtos);
1.930 + __ pop_i(rbx);
1.931 + // rax,: value
1.932 + // rdx: array
1.933 + index_check(rdx, rbx); // prefer index in rbx,
1.934 + // rbx,: index
1.935 + __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax);
1.936 +}
1.937 +
1.938 +
1.939 +void TemplateTable::sastore() {
1.940 + castore();
1.941 +}
1.942 +
1.943 +
1.944 +void TemplateTable::istore(int n) {
1.945 + transition(itos, vtos);
1.946 + __ movl(iaddress(n), rax);
1.947 + __ tag_local(frame::TagValue, n);
1.948 +}
1.949 +
1.950 +
1.951 +void TemplateTable::lstore(int n) {
1.952 + transition(ltos, vtos);
1.953 + __ movl(laddress(n), rax);
1.954 + __ movl(haddress(n), rdx);
1.955 + __ tag_local(frame::TagCategory2, n);
1.956 +}
1.957 +
1.958 +
1.959 +void TemplateTable::fstore(int n) {
1.960 + transition(ftos, vtos);
1.961 + __ fstp_s(faddress(n));
1.962 + __ tag_local(frame::TagValue, n);
1.963 +}
1.964 +
1.965 +
1.966 +void TemplateTable::dstore(int n) {
1.967 + transition(dtos, vtos);
1.968 + if (TaggedStackInterpreter) {
1.969 + __ subl(rsp, 2 * wordSize);
1.970 + __ fstp_d(Address(rsp, 0));
1.971 + __ popl(rax);
1.972 + __ popl(rdx);
1.973 + __ movl(laddress(n), rax);
1.974 + __ movl(haddress(n), rdx);
1.975 + __ tag_local(frame::TagCategory2, n);
1.976 + } else {
1.977 + __ fstp_d(daddress(n));
1.978 + }
1.979 +}
1.980 +
1.981 +
1.982 +void TemplateTable::astore(int n) {
1.983 + transition(vtos, vtos);
1.984 + __ pop_ptr(rax, rdx);
1.985 + __ movl(aaddress(n), rax);
1.986 + __ tag_local(rdx, n);
1.987 +}
1.988 +
1.989 +
1.990 +void TemplateTable::pop() {
1.991 + transition(vtos, vtos);
1.992 + __ addl(rsp, Interpreter::stackElementSize());
1.993 +}
1.994 +
1.995 +
1.996 +void TemplateTable::pop2() {
1.997 + transition(vtos, vtos);
1.998 + __ addl(rsp, 2*Interpreter::stackElementSize());
1.999 +}
1.1000 +
1.1001 +
1.1002 +void TemplateTable::dup() {
1.1003 + transition(vtos, vtos);
1.1004 + // stack: ..., a
1.1005 + __ load_ptr_and_tag(0, rax, rdx);
1.1006 + __ push_ptr(rax, rdx);
1.1007 + // stack: ..., a, a
1.1008 +}
1.1009 +
1.1010 +
1.1011 +void TemplateTable::dup_x1() {
1.1012 + transition(vtos, vtos);
1.1013 + // stack: ..., a, b
1.1014 + __ load_ptr_and_tag(0, rax, rdx); // load b
1.1015 + __ load_ptr_and_tag(1, rcx, rbx); // load a
1.1016 + __ store_ptr_and_tag(1, rax, rdx); // store b
1.1017 + __ store_ptr_and_tag(0, rcx, rbx); // store a
1.1018 + __ push_ptr(rax, rdx); // push b
1.1019 + // stack: ..., b, a, b
1.1020 +}
1.1021 +
1.1022 +
1.1023 +void TemplateTable::dup_x2() {
1.1024 + transition(vtos, vtos);
1.1025 + // stack: ..., a, b, c
1.1026 + __ load_ptr_and_tag(0, rax, rdx); // load c
1.1027 + __ load_ptr_and_tag(2, rcx, rbx); // load a
1.1028 + __ store_ptr_and_tag(2, rax, rdx); // store c in a
1.1029 + __ push_ptr(rax, rdx); // push c
1.1030 + // stack: ..., c, b, c, c
1.1031 + __ load_ptr_and_tag(2, rax, rdx); // load b
1.1032 + __ store_ptr_and_tag(2, rcx, rbx); // store a in b
1.1033 + // stack: ..., c, a, c, c
1.1034 + __ store_ptr_and_tag(1, rax, rdx); // store b in c
1.1035 + // stack: ..., c, a, b, c
1.1036 +}
1.1037 +
1.1038 +
1.1039 +void TemplateTable::dup2() {
1.1040 + transition(vtos, vtos);
1.1041 + // stack: ..., a, b
1.1042 + __ load_ptr_and_tag(1, rax, rdx); // load a
1.1043 + __ push_ptr(rax, rdx); // push a
1.1044 + __ load_ptr_and_tag(1, rax, rdx); // load b
1.1045 + __ push_ptr(rax, rdx); // push b
1.1046 + // stack: ..., a, b, a, b
1.1047 +}
1.1048 +
1.1049 +
1.1050 +void TemplateTable::dup2_x1() {
1.1051 + transition(vtos, vtos);
1.1052 + // stack: ..., a, b, c
1.1053 + __ load_ptr_and_tag(0, rcx, rbx); // load c
1.1054 + __ load_ptr_and_tag(1, rax, rdx); // load b
1.1055 + __ push_ptr(rax, rdx); // push b
1.1056 + __ push_ptr(rcx, rbx); // push c
1.1057 + // stack: ..., a, b, c, b, c
1.1058 + __ store_ptr_and_tag(3, rcx, rbx); // store c in b
1.1059 + // stack: ..., a, c, c, b, c
1.1060 + __ load_ptr_and_tag(4, rcx, rbx); // load a
1.1061 + __ store_ptr_and_tag(2, rcx, rbx); // store a in 2nd c
1.1062 + // stack: ..., a, c, a, b, c
1.1063 + __ store_ptr_and_tag(4, rax, rdx); // store b in a
1.1064 + // stack: ..., b, c, a, b, c
1.1065 + // stack: ..., b, c, a, b, c
1.1066 +}
1.1067 +
1.1068 +
1.1069 +void TemplateTable::dup2_x2() {
1.1070 + transition(vtos, vtos);
1.1071 + // stack: ..., a, b, c, d
1.1072 + __ load_ptr_and_tag(0, rcx, rbx); // load d
1.1073 + __ load_ptr_and_tag(1, rax, rdx); // load c
1.1074 + __ push_ptr(rax, rdx); // push c
1.1075 + __ push_ptr(rcx, rbx); // push d
1.1076 + // stack: ..., a, b, c, d, c, d
1.1077 + __ load_ptr_and_tag(4, rax, rdx); // load b
1.1078 + __ store_ptr_and_tag(2, rax, rdx); // store b in d
1.1079 + __ store_ptr_and_tag(4, rcx, rbx); // store d in b
1.1080 + // stack: ..., a, d, c, b, c, d
1.1081 + __ load_ptr_and_tag(5, rcx, rbx); // load a
1.1082 + __ load_ptr_and_tag(3, rax, rdx); // load c
1.1083 + __ store_ptr_and_tag(3, rcx, rbx); // store a in c
1.1084 + __ store_ptr_and_tag(5, rax, rdx); // store c in a
1.1085 + // stack: ..., c, d, a, b, c, d
1.1086 + // stack: ..., c, d, a, b, c, d
1.1087 +}
1.1088 +
1.1089 +
1.1090 +void TemplateTable::swap() {
1.1091 + transition(vtos, vtos);
1.1092 + // stack: ..., a, b
1.1093 + __ load_ptr_and_tag(1, rcx, rbx); // load a
1.1094 + __ load_ptr_and_tag(0, rax, rdx); // load b
1.1095 + __ store_ptr_and_tag(0, rcx, rbx); // store a in b
1.1096 + __ store_ptr_and_tag(1, rax, rdx); // store b in a
1.1097 + // stack: ..., b, a
1.1098 +}
1.1099 +
1.1100 +
1.1101 +void TemplateTable::iop2(Operation op) {
1.1102 + transition(itos, itos);
1.1103 + switch (op) {
1.1104 + case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1.1105 + case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1.1106 + case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1.1107 + case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1.1108 + case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1.1109 + case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1.1110 + case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1.1111 + case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1.1112 + case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1.1113 + default : ShouldNotReachHere();
1.1114 + }
1.1115 +}
1.1116 +
1.1117 +
1.1118 +void TemplateTable::lop2(Operation op) {
1.1119 + transition(ltos, ltos);
1.1120 + __ pop_l(rbx, rcx);
1.1121 + switch (op) {
1.1122 + case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1.1123 + case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1.1124 + __ movl(rax, rbx); __ movl(rdx, rcx); break;
1.1125 + case _and: __ andl(rax, rbx); __ andl(rdx, rcx); break;
1.1126 + case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1.1127 + case _xor: __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1.1128 + default : ShouldNotReachHere();
1.1129 + }
1.1130 +}
1.1131 +
1.1132 +
1.1133 +void TemplateTable::idiv() {
1.1134 + transition(itos, itos);
1.1135 + __ movl(rcx, rax);
1.1136 + __ pop_i(rax);
1.1137 + // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1.1138 + // they are not equal, one could do a normal division (no correction
1.1139 + // needed), which may speed up this implementation for the common case.
1.1140 + // (see also JVM spec., p.243 & p.271)
1.1141 + __ corrected_idivl(rcx);
1.1142 +}
1.1143 +
1.1144 +
1.1145 +void TemplateTable::irem() {
1.1146 + transition(itos, itos);
1.1147 + __ movl(rcx, rax);
1.1148 + __ pop_i(rax);
1.1149 + // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1.1150 + // they are not equal, one could do a normal division (no correction
1.1151 + // needed), which may speed up this implementation for the common case.
1.1152 + // (see also JVM spec., p.243 & p.271)
1.1153 + __ corrected_idivl(rcx);
1.1154 + __ movl(rax, rdx);
1.1155 +}
1.1156 +
1.1157 +
1.1158 +void TemplateTable::lmul() {
1.1159 + transition(ltos, ltos);
1.1160 + __ pop_l(rbx, rcx);
1.1161 + __ pushl(rcx); __ pushl(rbx);
1.1162 + __ pushl(rdx); __ pushl(rax);
1.1163 + __ lmul(2 * wordSize, 0);
1.1164 + __ addl(rsp, 4 * wordSize); // take off temporaries
1.1165 +}
1.1166 +
1.1167 +
1.1168 +void TemplateTable::ldiv() {
1.1169 + transition(ltos, ltos);
1.1170 + __ pop_l(rbx, rcx);
1.1171 + __ pushl(rcx); __ pushl(rbx);
1.1172 + __ pushl(rdx); __ pushl(rax);
1.1173 + // check if y = 0
1.1174 + __ orl(rax, rdx);
1.1175 + __ jump_cc(Assembler::zero,
1.1176 + ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1.1177 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1.1178 + __ addl(rsp, 4 * wordSize); // take off temporaries
1.1179 +}
1.1180 +
1.1181 +
1.1182 +void TemplateTable::lrem() {
1.1183 + transition(ltos, ltos);
1.1184 + __ pop_l(rbx, rcx);
1.1185 + __ pushl(rcx); __ pushl(rbx);
1.1186 + __ pushl(rdx); __ pushl(rax);
1.1187 + // check if y = 0
1.1188 + __ orl(rax, rdx);
1.1189 + __ jump_cc(Assembler::zero,
1.1190 + ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1.1191 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1.1192 + __ addl(rsp, 4 * wordSize);
1.1193 +}
1.1194 +
1.1195 +
1.1196 +void TemplateTable::lshl() {
1.1197 + transition(itos, ltos);
1.1198 + __ movl(rcx, rax); // get shift count
1.1199 + __ pop_l(rax, rdx); // get shift value
1.1200 + __ lshl(rdx, rax);
1.1201 +}
1.1202 +
1.1203 +
1.1204 +void TemplateTable::lshr() {
1.1205 + transition(itos, ltos);
1.1206 + __ movl(rcx, rax); // get shift count
1.1207 + __ pop_l(rax, rdx); // get shift value
1.1208 + __ lshr(rdx, rax, true);
1.1209 +}
1.1210 +
1.1211 +
1.1212 +void TemplateTable::lushr() {
1.1213 + transition(itos, ltos);
1.1214 + __ movl(rcx, rax); // get shift count
1.1215 + __ pop_l(rax, rdx); // get shift value
1.1216 + __ lshr(rdx, rax);
1.1217 +}
1.1218 +
1.1219 +
1.1220 +void TemplateTable::fop2(Operation op) {
1.1221 + transition(ftos, ftos);
1.1222 + __ pop_ftos_to_rsp(); // pop ftos into rsp
1.1223 + switch (op) {
1.1224 + case add: __ fadd_s (at_rsp()); break;
1.1225 + case sub: __ fsubr_s(at_rsp()); break;
1.1226 + case mul: __ fmul_s (at_rsp()); break;
1.1227 + case div: __ fdivr_s(at_rsp()); break;
1.1228 + case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1.1229 + default : ShouldNotReachHere();
1.1230 + }
1.1231 + __ f2ieee();
1.1232 + __ popl(rax); // pop float thing off
1.1233 +}
1.1234 +
1.1235 +
1.1236 +void TemplateTable::dop2(Operation op) {
1.1237 + transition(dtos, dtos);
1.1238 + __ pop_dtos_to_rsp(); // pop dtos into rsp
1.1239 +
1.1240 + switch (op) {
1.1241 + case add: __ fadd_d (at_rsp()); break;
1.1242 + case sub: __ fsubr_d(at_rsp()); break;
1.1243 + case mul: {
1.1244 + Label L_strict;
1.1245 + Label L_join;
1.1246 + const Address access_flags (rcx, methodOopDesc::access_flags_offset());
1.1247 + __ get_method(rcx);
1.1248 + __ movl(rcx, access_flags);
1.1249 + __ testl(rcx, JVM_ACC_STRICT);
1.1250 + __ jccb(Assembler::notZero, L_strict);
1.1251 + __ fmul_d (at_rsp());
1.1252 + __ jmpb(L_join);
1.1253 + __ bind(L_strict);
1.1254 + __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1.1255 + __ fmulp();
1.1256 + __ fmul_d (at_rsp());
1.1257 + __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1.1258 + __ fmulp();
1.1259 + __ bind(L_join);
1.1260 + break;
1.1261 + }
1.1262 + case div: {
1.1263 + Label L_strict;
1.1264 + Label L_join;
1.1265 + const Address access_flags (rcx, methodOopDesc::access_flags_offset());
1.1266 + __ get_method(rcx);
1.1267 + __ movl(rcx, access_flags);
1.1268 + __ testl(rcx, JVM_ACC_STRICT);
1.1269 + __ jccb(Assembler::notZero, L_strict);
1.1270 + __ fdivr_d(at_rsp());
1.1271 + __ jmp(L_join);
1.1272 + __ bind(L_strict);
1.1273 + __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1.1274 + __ fmul_d (at_rsp());
1.1275 + __ fdivrp();
1.1276 + __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1.1277 + __ fmulp();
1.1278 + __ bind(L_join);
1.1279 + break;
1.1280 + }
1.1281 + case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1.1282 + default : ShouldNotReachHere();
1.1283 + }
1.1284 + __ d2ieee();
1.1285 + // Pop double precision number from rsp.
1.1286 + __ popl(rax);
1.1287 + __ popl(rdx);
1.1288 +}
1.1289 +
1.1290 +
1.1291 +void TemplateTable::ineg() {
1.1292 + transition(itos, itos);
1.1293 + __ negl(rax);
1.1294 +}
1.1295 +
1.1296 +
1.1297 +void TemplateTable::lneg() {
1.1298 + transition(ltos, ltos);
1.1299 + __ lneg(rdx, rax);
1.1300 +}
1.1301 +
1.1302 +
1.1303 +void TemplateTable::fneg() {
1.1304 + transition(ftos, ftos);
1.1305 + __ fchs();
1.1306 +}
1.1307 +
1.1308 +
1.1309 +void TemplateTable::dneg() {
1.1310 + transition(dtos, dtos);
1.1311 + __ fchs();
1.1312 +}
1.1313 +
1.1314 +
1.1315 +void TemplateTable::iinc() {
1.1316 + transition(vtos, vtos);
1.1317 + __ load_signed_byte(rdx, at_bcp(2)); // get constant
1.1318 + locals_index(rbx);
1.1319 + __ addl(iaddress(rbx), rdx);
1.1320 +}
1.1321 +
1.1322 +
1.1323 +void TemplateTable::wide_iinc() {
1.1324 + transition(vtos, vtos);
1.1325 + __ movl(rdx, at_bcp(4)); // get constant
1.1326 + locals_index_wide(rbx);
1.1327 + __ bswap(rdx); // swap bytes & sign-extend constant
1.1328 + __ sarl(rdx, 16);
1.1329 + __ addl(iaddress(rbx), rdx);
1.1330 + // Note: should probably use only one movl to get both
1.1331 + // the index and the constant -> fix this
1.1332 +}
1.1333 +
1.1334 +
1.1335 +void TemplateTable::convert() {
1.1336 + // Checking
1.1337 +#ifdef ASSERT
1.1338 + { TosState tos_in = ilgl;
1.1339 + TosState tos_out = ilgl;
1.1340 + switch (bytecode()) {
1.1341 + case Bytecodes::_i2l: // fall through
1.1342 + case Bytecodes::_i2f: // fall through
1.1343 + case Bytecodes::_i2d: // fall through
1.1344 + case Bytecodes::_i2b: // fall through
1.1345 + case Bytecodes::_i2c: // fall through
1.1346 + case Bytecodes::_i2s: tos_in = itos; break;
1.1347 + case Bytecodes::_l2i: // fall through
1.1348 + case Bytecodes::_l2f: // fall through
1.1349 + case Bytecodes::_l2d: tos_in = ltos; break;
1.1350 + case Bytecodes::_f2i: // fall through
1.1351 + case Bytecodes::_f2l: // fall through
1.1352 + case Bytecodes::_f2d: tos_in = ftos; break;
1.1353 + case Bytecodes::_d2i: // fall through
1.1354 + case Bytecodes::_d2l: // fall through
1.1355 + case Bytecodes::_d2f: tos_in = dtos; break;
1.1356 + default : ShouldNotReachHere();
1.1357 + }
1.1358 + switch (bytecode()) {
1.1359 + case Bytecodes::_l2i: // fall through
1.1360 + case Bytecodes::_f2i: // fall through
1.1361 + case Bytecodes::_d2i: // fall through
1.1362 + case Bytecodes::_i2b: // fall through
1.1363 + case Bytecodes::_i2c: // fall through
1.1364 + case Bytecodes::_i2s: tos_out = itos; break;
1.1365 + case Bytecodes::_i2l: // fall through
1.1366 + case Bytecodes::_f2l: // fall through
1.1367 + case Bytecodes::_d2l: tos_out = ltos; break;
1.1368 + case Bytecodes::_i2f: // fall through
1.1369 + case Bytecodes::_l2f: // fall through
1.1370 + case Bytecodes::_d2f: tos_out = ftos; break;
1.1371 + case Bytecodes::_i2d: // fall through
1.1372 + case Bytecodes::_l2d: // fall through
1.1373 + case Bytecodes::_f2d: tos_out = dtos; break;
1.1374 + default : ShouldNotReachHere();
1.1375 + }
1.1376 + transition(tos_in, tos_out);
1.1377 + }
1.1378 +#endif // ASSERT
1.1379 +
1.1380 + // Conversion
1.1381 + // (Note: use pushl(rcx)/popl(rcx) for 1/2-word stack-ptr manipulation)
1.1382 + switch (bytecode()) {
1.1383 + case Bytecodes::_i2l:
1.1384 + __ extend_sign(rdx, rax);
1.1385 + break;
1.1386 + case Bytecodes::_i2f:
1.1387 + __ pushl(rax); // store int on tos
1.1388 + __ fild_s(at_rsp()); // load int to ST0
1.1389 + __ f2ieee(); // truncate to float size
1.1390 + __ popl(rcx); // adjust rsp
1.1391 + break;
1.1392 + case Bytecodes::_i2d:
1.1393 + __ pushl(rax); // add one slot for d2ieee()
1.1394 + __ pushl(rax); // store int on tos
1.1395 + __ fild_s(at_rsp()); // load int to ST0
1.1396 + __ d2ieee(); // truncate to double size
1.1397 + __ popl(rcx); // adjust rsp
1.1398 + __ popl(rcx);
1.1399 + break;
1.1400 + case Bytecodes::_i2b:
1.1401 + __ shll(rax, 24); // truncate upper 24 bits
1.1402 + __ sarl(rax, 24); // and sign-extend byte
1.1403 + break;
1.1404 + case Bytecodes::_i2c:
1.1405 + __ andl(rax, 0xFFFF); // truncate upper 16 bits
1.1406 + break;
1.1407 + case Bytecodes::_i2s:
1.1408 + __ shll(rax, 16); // truncate upper 16 bits
1.1409 + __ sarl(rax, 16); // and sign-extend short
1.1410 + break;
1.1411 + case Bytecodes::_l2i:
1.1412 + /* nothing to do */
1.1413 + break;
1.1414 + case Bytecodes::_l2f:
1.1415 + __ pushl(rdx); // store long on tos
1.1416 + __ pushl(rax);
1.1417 + __ fild_d(at_rsp()); // load long to ST0
1.1418 + __ f2ieee(); // truncate to float size
1.1419 + __ popl(rcx); // adjust rsp
1.1420 + __ popl(rcx);
1.1421 + break;
1.1422 + case Bytecodes::_l2d:
1.1423 + __ pushl(rdx); // store long on tos
1.1424 + __ pushl(rax);
1.1425 + __ fild_d(at_rsp()); // load long to ST0
1.1426 + __ d2ieee(); // truncate to double size
1.1427 + __ popl(rcx); // adjust rsp
1.1428 + __ popl(rcx);
1.1429 + break;
1.1430 + case Bytecodes::_f2i:
1.1431 + __ pushl(rcx); // reserve space for argument
1.1432 + __ fstp_s(at_rsp()); // pass float argument on stack
1.1433 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1.1434 + break;
1.1435 + case Bytecodes::_f2l:
1.1436 + __ pushl(rcx); // reserve space for argument
1.1437 + __ fstp_s(at_rsp()); // pass float argument on stack
1.1438 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1.1439 + break;
1.1440 + case Bytecodes::_f2d:
1.1441 + /* nothing to do */
1.1442 + break;
1.1443 + case Bytecodes::_d2i:
1.1444 + __ pushl(rcx); // reserve space for argument
1.1445 + __ pushl(rcx);
1.1446 + __ fstp_d(at_rsp()); // pass double argument on stack
1.1447 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1.1448 + break;
1.1449 + case Bytecodes::_d2l:
1.1450 + __ pushl(rcx); // reserve space for argument
1.1451 + __ pushl(rcx);
1.1452 + __ fstp_d(at_rsp()); // pass double argument on stack
1.1453 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1.1454 + break;
1.1455 + case Bytecodes::_d2f:
1.1456 + __ pushl(rcx); // reserve space for f2ieee()
1.1457 + __ f2ieee(); // truncate to float size
1.1458 + __ popl(rcx); // adjust rsp
1.1459 + break;
1.1460 + default :
1.1461 + ShouldNotReachHere();
1.1462 + }
1.1463 +}
1.1464 +
1.1465 +
1.1466 +void TemplateTable::lcmp() {
1.1467 + transition(ltos, itos);
1.1468 + // y = rdx:rax
1.1469 + __ pop_l(rbx, rcx); // get x = rcx:rbx
1.1470 + __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1.1471 + __ movl(rax, rcx);
1.1472 +}
1.1473 +
1.1474 +
1.1475 +void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1.1476 + if (is_float) {
1.1477 + __ pop_ftos_to_rsp();
1.1478 + __ fld_s(at_rsp());
1.1479 + } else {
1.1480 + __ pop_dtos_to_rsp();
1.1481 + __ fld_d(at_rsp());
1.1482 + __ popl(rdx);
1.1483 + }
1.1484 + __ popl(rcx);
1.1485 + __ fcmp2int(rax, unordered_result < 0);
1.1486 +}
1.1487 +
1.1488 +
1.1489 +void TemplateTable::branch(bool is_jsr, bool is_wide) {
1.1490 + __ get_method(rcx); // ECX holds method
1.1491 + __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count
1.1492 +
1.1493 + const ByteSize be_offset = methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset();
1.1494 + const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset();
1.1495 + const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1.1496 +
1.1497 + // Load up EDX with the branch displacement
1.1498 + __ movl(rdx, at_bcp(1));
1.1499 + __ bswap(rdx);
1.1500 + if (!is_wide) __ sarl(rdx, 16);
1.1501 +
1.1502 + // Handle all the JSR stuff here, then exit.
1.1503 + // It's much shorter and cleaner than intermingling with the
1.1504 + // non-JSR normal-branch stuff occuring below.
1.1505 + if (is_jsr) {
1.1506 + // Pre-load the next target bytecode into EBX
1.1507 + __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0));
1.1508 +
1.1509 + // compute return address as bci in rax,
1.1510 + __ leal(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(constMethodOopDesc::codes_offset())));
1.1511 + __ subl(rax, Address(rcx, methodOopDesc::const_offset()));
1.1512 + // Adjust the bcp in ESI by the displacement in EDX
1.1513 + __ addl(rsi, rdx);
1.1514 + // Push return address
1.1515 + __ push_i(rax);
1.1516 + // jsr returns vtos
1.1517 + __ dispatch_only_noverify(vtos);
1.1518 + return;
1.1519 + }
1.1520 +
1.1521 + // Normal (non-jsr) branch handling
1.1522 +
1.1523 + // Adjust the bcp in ESI by the displacement in EDX
1.1524 + __ addl(rsi, rdx);
1.1525 +
1.1526 + assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters");
1.1527 + Label backedge_counter_overflow;
1.1528 + Label profile_method;
1.1529 + Label dispatch;
1.1530 + if (UseLoopCounter) {
1.1531 + // increment backedge counter for backward branches
1.1532 + // rax,: MDO
1.1533 + // rbx,: MDO bumped taken-count
1.1534 + // rcx: method
1.1535 + // rdx: target offset
1.1536 + // rsi: target bcp
1.1537 + // rdi: locals pointer
1.1538 + __ testl(rdx, rdx); // check if forward or backward branch
1.1539 + __ jcc(Assembler::positive, dispatch); // count only if backward branch
1.1540 +
1.1541 + // increment counter
1.1542 + __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1.1543 + __ increment(rax, InvocationCounter::count_increment); // increment counter
1.1544 + __ movl(Address(rcx, be_offset), rax); // store counter
1.1545 +
1.1546 + __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1.1547 + __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1.1548 + __ addl(rax, Address(rcx, be_offset)); // add both counters
1.1549 +
1.1550 + if (ProfileInterpreter) {
1.1551 + // Test to see if we should create a method data oop
1.1552 + __ cmp32(rax,
1.1553 + ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1.1554 + __ jcc(Assembler::less, dispatch);
1.1555 +
1.1556 + // if no method data exists, go to profile method
1.1557 + __ test_method_data_pointer(rax, profile_method);
1.1558 +
1.1559 + if (UseOnStackReplacement) {
1.1560 + // check for overflow against rbx, which is the MDO taken count
1.1561 + __ cmp32(rbx,
1.1562 + ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1.1563 + __ jcc(Assembler::below, dispatch);
1.1564 +
1.1565 + // When ProfileInterpreter is on, the backedge_count comes from the
1.1566 + // methodDataOop, which value does not get reset on the call to
1.1567 + // frequency_counter_overflow(). To avoid excessive calls to the overflow
1.1568 + // routine while the method is being compiled, add a second test to make
1.1569 + // sure the overflow function is called only once every overflow_frequency.
1.1570 + const int overflow_frequency = 1024;
1.1571 + __ andl(rbx, overflow_frequency-1);
1.1572 + __ jcc(Assembler::zero, backedge_counter_overflow);
1.1573 +
1.1574 + }
1.1575 + } else {
1.1576 + if (UseOnStackReplacement) {
1.1577 + // check for overflow against rax, which is the sum of the counters
1.1578 + __ cmp32(rax,
1.1579 + ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1.1580 + __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1.1581 +
1.1582 + }
1.1583 + }
1.1584 + __ bind(dispatch);
1.1585 + }
1.1586 +
1.1587 + // Pre-load the next target bytecode into EBX
1.1588 + __ load_unsigned_byte(rbx, Address(rsi, 0));
1.1589 +
1.1590 + // continue with the bytecode @ target
1.1591 + // rax,: return bci for jsr's, unused otherwise
1.1592 + // rbx,: target bytecode
1.1593 + // rsi: target bcp
1.1594 + __ dispatch_only(vtos);
1.1595 +
1.1596 + if (UseLoopCounter) {
1.1597 + if (ProfileInterpreter) {
1.1598 + // Out-of-line code to allocate method data oop.
1.1599 + __ bind(profile_method);
1.1600 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi);
1.1601 + __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1.1602 + __ movl(rcx, Address(rbp, method_offset));
1.1603 + __ movl(rcx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
1.1604 + __ movl(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
1.1605 + __ test_method_data_pointer(rcx, dispatch);
1.1606 + // offset non-null mdp by MDO::data_offset() + IR::profile_method()
1.1607 + __ addl(rcx, in_bytes(methodDataOopDesc::data_offset()));
1.1608 + __ addl(rcx, rax);
1.1609 + __ movl(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
1.1610 + __ jmp(dispatch);
1.1611 + }
1.1612 +
1.1613 + if (UseOnStackReplacement) {
1.1614 +
1.1615 + // invocation counter overflow
1.1616 + __ bind(backedge_counter_overflow);
1.1617 + __ negl(rdx);
1.1618 + __ addl(rdx, rsi); // branch bcp
1.1619 + call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx);
1.1620 + __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1.1621 +
1.1622 + // rax,: osr nmethod (osr ok) or NULL (osr not possible)
1.1623 + // rbx,: target bytecode
1.1624 + // rdx: scratch
1.1625 + // rdi: locals pointer
1.1626 + // rsi: bcp
1.1627 + __ testl(rax, rax); // test result
1.1628 + __ jcc(Assembler::zero, dispatch); // no osr if null
1.1629 + // nmethod may have been invalidated (VM may block upon call_VM return)
1.1630 + __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1.1631 + __ cmpl(rcx, InvalidOSREntryBci);
1.1632 + __ jcc(Assembler::equal, dispatch);
1.1633 +
1.1634 + // We have the address of an on stack replacement routine in rax,
1.1635 + // We need to prepare to execute the OSR method. First we must
1.1636 + // migrate the locals and monitors off of the stack.
1.1637 +
1.1638 + __ movl(rsi, rax); // save the nmethod
1.1639 +
1.1640 + const Register thread = rcx;
1.1641 + __ get_thread(thread);
1.1642 + call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1.1643 + // rax, is OSR buffer, move it to expected parameter location
1.1644 + __ movl(rcx, rax);
1.1645 +
1.1646 + // pop the interpreter frame
1.1647 + __ movl(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1.1648 + __ leave(); // remove frame anchor
1.1649 + __ popl(rdi); // get return address
1.1650 + __ movl(rsp, rdx); // set sp to sender sp
1.1651 +
1.1652 +
1.1653 + Label skip;
1.1654 + Label chkint;
1.1655 +
1.1656 + // The interpreter frame we have removed may be returning to
1.1657 + // either the callstub or the interpreter. Since we will
1.1658 + // now be returning from a compiled (OSR) nmethod we must
1.1659 + // adjust the return to the return were it can handler compiled
1.1660 + // results and clean the fpu stack. This is very similar to
1.1661 + // what a i2c adapter must do.
1.1662 +
1.1663 + // Are we returning to the call stub?
1.1664 +
1.1665 + __ cmp32(rdi, ExternalAddress(StubRoutines::_call_stub_return_address));
1.1666 + __ jcc(Assembler::notEqual, chkint);
1.1667 +
1.1668 + // yes adjust to the specialized call stub return.
1.1669 + assert(StubRoutines::i486::get_call_stub_compiled_return() != NULL, "must be set");
1.1670 + __ lea(rdi, ExternalAddress(StubRoutines::i486::get_call_stub_compiled_return()));
1.1671 + __ jmp(skip);
1.1672 +
1.1673 + __ bind(chkint);
1.1674 +
1.1675 + // Are we returning to the interpreter? Look for sentinel
1.1676 +
1.1677 + __ cmpl(Address(rdi, -8), Interpreter::return_sentinel);
1.1678 + __ jcc(Assembler::notEqual, skip);
1.1679 +
1.1680 + // Adjust to compiled return back to interpreter
1.1681 +
1.1682 + __ movl(rdi, Address(rdi, -4));
1.1683 + __ bind(skip);
1.1684 +
1.1685 + // Align stack pointer for compiled code (note that caller is
1.1686 + // responsible for undoing this fixup by remembering the old SP
1.1687 + // in an rbp,-relative location)
1.1688 + __ andl(rsp, -(StackAlignmentInBytes));
1.1689 +
1.1690 + // push the (possibly adjusted) return address
1.1691 + __ pushl(rdi);
1.1692 +
1.1693 + // and begin the OSR nmethod
1.1694 + __ jmp(Address(rsi, nmethod::osr_entry_point_offset()));
1.1695 + }
1.1696 + }
1.1697 +}
1.1698 +
1.1699 +
1.1700 +void TemplateTable::if_0cmp(Condition cc) {
1.1701 + transition(itos, vtos);
1.1702 + // assume branch is more often taken than not (loops use backward branches)
1.1703 + Label not_taken;
1.1704 + __ testl(rax, rax);
1.1705 + __ jcc(j_not(cc), not_taken);
1.1706 + branch(false, false);
1.1707 + __ bind(not_taken);
1.1708 + __ profile_not_taken_branch(rax);
1.1709 +}
1.1710 +
1.1711 +
1.1712 +void TemplateTable::if_icmp(Condition cc) {
1.1713 + transition(itos, vtos);
1.1714 + // assume branch is more often taken than not (loops use backward branches)
1.1715 + Label not_taken;
1.1716 + __ pop_i(rdx);
1.1717 + __ cmpl(rdx, rax);
1.1718 + __ jcc(j_not(cc), not_taken);
1.1719 + branch(false, false);
1.1720 + __ bind(not_taken);
1.1721 + __ profile_not_taken_branch(rax);
1.1722 +}
1.1723 +
1.1724 +
1.1725 +void TemplateTable::if_nullcmp(Condition cc) {
1.1726 + transition(atos, vtos);
1.1727 + // assume branch is more often taken than not (loops use backward branches)
1.1728 + Label not_taken;
1.1729 + __ testl(rax, rax);
1.1730 + __ jcc(j_not(cc), not_taken);
1.1731 + branch(false, false);
1.1732 + __ bind(not_taken);
1.1733 + __ profile_not_taken_branch(rax);
1.1734 +}
1.1735 +
1.1736 +
1.1737 +void TemplateTable::if_acmp(Condition cc) {
1.1738 + transition(atos, vtos);
1.1739 + // assume branch is more often taken than not (loops use backward branches)
1.1740 + Label not_taken;
1.1741 + __ pop_ptr(rdx);
1.1742 + __ cmpl(rdx, rax);
1.1743 + __ jcc(j_not(cc), not_taken);
1.1744 + branch(false, false);
1.1745 + __ bind(not_taken);
1.1746 + __ profile_not_taken_branch(rax);
1.1747 +}
1.1748 +
1.1749 +
1.1750 +void TemplateTable::ret() {
1.1751 + transition(vtos, vtos);
1.1752 + locals_index(rbx);
1.1753 + __ movl(rbx, iaddress(rbx)); // get return bci, compute return bcp
1.1754 + __ profile_ret(rbx, rcx);
1.1755 + __ get_method(rax);
1.1756 + __ movl(rsi, Address(rax, methodOopDesc::const_offset()));
1.1757 + __ leal(rsi, Address(rsi, rbx, Address::times_1,
1.1758 + constMethodOopDesc::codes_offset()));
1.1759 + __ dispatch_next(vtos);
1.1760 +}
1.1761 +
1.1762 +
1.1763 +void TemplateTable::wide_ret() {
1.1764 + transition(vtos, vtos);
1.1765 + locals_index_wide(rbx);
1.1766 + __ movl(rbx, iaddress(rbx)); // get return bci, compute return bcp
1.1767 + __ profile_ret(rbx, rcx);
1.1768 + __ get_method(rax);
1.1769 + __ movl(rsi, Address(rax, methodOopDesc::const_offset()));
1.1770 + __ leal(rsi, Address(rsi, rbx, Address::times_1, constMethodOopDesc::codes_offset()));
1.1771 + __ dispatch_next(vtos);
1.1772 +}
1.1773 +
1.1774 +
1.1775 +void TemplateTable::tableswitch() {
1.1776 + Label default_case, continue_execution;
1.1777 + transition(itos, vtos);
1.1778 + // align rsi
1.1779 + __ leal(rbx, at_bcp(wordSize));
1.1780 + __ andl(rbx, -wordSize);
1.1781 + // load lo & hi
1.1782 + __ movl(rcx, Address(rbx, 1 * wordSize));
1.1783 + __ movl(rdx, Address(rbx, 2 * wordSize));
1.1784 + __ bswap(rcx);
1.1785 + __ bswap(rdx);
1.1786 + // check against lo & hi
1.1787 + __ cmpl(rax, rcx);
1.1788 + __ jccb(Assembler::less, default_case);
1.1789 + __ cmpl(rax, rdx);
1.1790 + __ jccb(Assembler::greater, default_case);
1.1791 + // lookup dispatch offset
1.1792 + __ subl(rax, rcx);
1.1793 + __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * wordSize));
1.1794 + __ profile_switch_case(rax, rbx, rcx);
1.1795 + // continue execution
1.1796 + __ bind(continue_execution);
1.1797 + __ bswap(rdx);
1.1798 + __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1.1799 + __ addl(rsi, rdx);
1.1800 + __ dispatch_only(vtos);
1.1801 + // handle default
1.1802 + __ bind(default_case);
1.1803 + __ profile_switch_default(rax);
1.1804 + __ movl(rdx, Address(rbx, 0));
1.1805 + __ jmp(continue_execution);
1.1806 +}
1.1807 +
1.1808 +
1.1809 +void TemplateTable::lookupswitch() {
1.1810 + transition(itos, itos);
1.1811 + __ stop("lookupswitch bytecode should have been rewritten");
1.1812 +}
1.1813 +
1.1814 +
1.1815 +void TemplateTable::fast_linearswitch() {
1.1816 + transition(itos, vtos);
1.1817 + Label loop_entry, loop, found, continue_execution;
1.1818 + // bswap rax, so we can avoid bswapping the table entries
1.1819 + __ bswap(rax);
1.1820 + // align rsi
1.1821 + __ leal(rbx, at_bcp(wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1.1822 + __ andl(rbx, -wordSize);
1.1823 + // set counter
1.1824 + __ movl(rcx, Address(rbx, wordSize));
1.1825 + __ bswap(rcx);
1.1826 + __ jmpb(loop_entry);
1.1827 + // table search
1.1828 + __ bind(loop);
1.1829 + __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize));
1.1830 + __ jccb(Assembler::equal, found);
1.1831 + __ bind(loop_entry);
1.1832 + __ decrement(rcx);
1.1833 + __ jcc(Assembler::greaterEqual, loop);
1.1834 + // default case
1.1835 + __ profile_switch_default(rax);
1.1836 + __ movl(rdx, Address(rbx, 0));
1.1837 + __ jmpb(continue_execution);
1.1838 + // entry found -> get offset
1.1839 + __ bind(found);
1.1840 + __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize));
1.1841 + __ profile_switch_case(rcx, rax, rbx);
1.1842 + // continue execution
1.1843 + __ bind(continue_execution);
1.1844 + __ bswap(rdx);
1.1845 + __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1.1846 + __ addl(rsi, rdx);
1.1847 + __ dispatch_only(vtos);
1.1848 +}
1.1849 +
1.1850 +
1.1851 +void TemplateTable::fast_binaryswitch() {
1.1852 + transition(itos, vtos);
1.1853 + // Implementation using the following core algorithm:
1.1854 + //
1.1855 + // int binary_search(int key, LookupswitchPair* array, int n) {
1.1856 + // // Binary search according to "Methodik des Programmierens" by
1.1857 + // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1.1858 + // int i = 0;
1.1859 + // int j = n;
1.1860 + // while (i+1 < j) {
1.1861 + // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1.1862 + // // with Q: for all i: 0 <= i < n: key < a[i]
1.1863 + // // where a stands for the array and assuming that the (inexisting)
1.1864 + // // element a[n] is infinitely big.
1.1865 + // int h = (i + j) >> 1;
1.1866 + // // i < h < j
1.1867 + // if (key < array[h].fast_match()) {
1.1868 + // j = h;
1.1869 + // } else {
1.1870 + // i = h;
1.1871 + // }
1.1872 + // }
1.1873 + // // R: a[i] <= key < a[i+1] or Q
1.1874 + // // (i.e., if key is within array, i is the correct index)
1.1875 + // return i;
1.1876 + // }
1.1877 +
1.1878 + // register allocation
1.1879 + const Register key = rax; // already set (tosca)
1.1880 + const Register array = rbx;
1.1881 + const Register i = rcx;
1.1882 + const Register j = rdx;
1.1883 + const Register h = rdi; // needs to be restored
1.1884 + const Register temp = rsi;
1.1885 + // setup array
1.1886 + __ save_bcp();
1.1887 +
1.1888 + __ leal(array, at_bcp(3*wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1.1889 + __ andl(array, -wordSize);
1.1890 + // initialize i & j
1.1891 + __ xorl(i, i); // i = 0;
1.1892 + __ movl(j, Address(array, -wordSize)); // j = length(array);
1.1893 + // Convert j into native byteordering
1.1894 + __ bswap(j);
1.1895 + // and start
1.1896 + Label entry;
1.1897 + __ jmp(entry);
1.1898 +
1.1899 + // binary search loop
1.1900 + { Label loop;
1.1901 + __ bind(loop);
1.1902 + // int h = (i + j) >> 1;
1.1903 + __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1.1904 + __ sarl(h, 1); // h = (i + j) >> 1;
1.1905 + // if (key < array[h].fast_match()) {
1.1906 + // j = h;
1.1907 + // } else {
1.1908 + // i = h;
1.1909 + // }
1.1910 + // Convert array[h].match to native byte-ordering before compare
1.1911 + __ movl(temp, Address(array, h, Address::times_8, 0*wordSize));
1.1912 + __ bswap(temp);
1.1913 + __ cmpl(key, temp);
1.1914 + if (VM_Version::supports_cmov()) {
1.1915 + __ cmovl(Assembler::less , j, h); // j = h if (key < array[h].fast_match())
1.1916 + __ cmovl(Assembler::greaterEqual, i, h); // i = h if (key >= array[h].fast_match())
1.1917 + } else {
1.1918 + Label set_i, end_of_if;
1.1919 + __ jccb(Assembler::greaterEqual, set_i); // {
1.1920 + __ movl(j, h); // j = h;
1.1921 + __ jmp(end_of_if); // }
1.1922 + __ bind(set_i); // else {
1.1923 + __ movl(i, h); // i = h;
1.1924 + __ bind(end_of_if); // }
1.1925 + }
1.1926 + // while (i+1 < j)
1.1927 + __ bind(entry);
1.1928 + __ leal(h, Address(i, 1)); // i+1
1.1929 + __ cmpl(h, j); // i+1 < j
1.1930 + __ jcc(Assembler::less, loop);
1.1931 + }
1.1932 +
1.1933 + // end of binary search, result index is i (must check again!)
1.1934 + Label default_case;
1.1935 + // Convert array[i].match to native byte-ordering before compare
1.1936 + __ movl(temp, Address(array, i, Address::times_8, 0*wordSize));
1.1937 + __ bswap(temp);
1.1938 + __ cmpl(key, temp);
1.1939 + __ jcc(Assembler::notEqual, default_case);
1.1940 +
1.1941 + // entry found -> j = offset
1.1942 + __ movl(j , Address(array, i, Address::times_8, 1*wordSize));
1.1943 + __ profile_switch_case(i, key, array);
1.1944 + __ bswap(j);
1.1945 + __ restore_bcp();
1.1946 + __ restore_locals(); // restore rdi
1.1947 + __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
1.1948 +
1.1949 + __ addl(rsi, j);
1.1950 + __ dispatch_only(vtos);
1.1951 +
1.1952 + // default case -> j = default offset
1.1953 + __ bind(default_case);
1.1954 + __ profile_switch_default(i);
1.1955 + __ movl(j, Address(array, -2*wordSize));
1.1956 + __ bswap(j);
1.1957 + __ restore_bcp();
1.1958 + __ restore_locals(); // restore rdi
1.1959 + __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
1.1960 + __ addl(rsi, j);
1.1961 + __ dispatch_only(vtos);
1.1962 +}
1.1963 +
1.1964 +
1.1965 +void TemplateTable::_return(TosState state) {
1.1966 + transition(state, state);
1.1967 + assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation
1.1968 +
1.1969 + if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
1.1970 + assert(state == vtos, "only valid state");
1.1971 + __ movl(rax, aaddress(0));
1.1972 + __ movl(rdi, Address(rax, oopDesc::klass_offset_in_bytes()));
1.1973 + __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
1.1974 + __ testl(rdi, JVM_ACC_HAS_FINALIZER);
1.1975 + Label skip_register_finalizer;
1.1976 + __ jcc(Assembler::zero, skip_register_finalizer);
1.1977 +
1.1978 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax);
1.1979 +
1.1980 + __ bind(skip_register_finalizer);
1.1981 + }
1.1982 +
1.1983 + __ remove_activation(state, rsi);
1.1984 + __ jmp(rsi);
1.1985 +}
1.1986 +
1.1987 +
1.1988 +// ----------------------------------------------------------------------------
1.1989 +// Volatile variables demand their effects be made known to all CPU's in
1.1990 +// order. Store buffers on most chips allow reads & writes to reorder; the
1.1991 +// JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
1.1992 +// memory barrier (i.e., it's not sufficient that the interpreter does not
1.1993 +// reorder volatile references, the hardware also must not reorder them).
1.1994 +//
1.1995 +// According to the new Java Memory Model (JMM):
1.1996 +// (1) All volatiles are serialized wrt to each other.
1.1997 +// ALSO reads & writes act as aquire & release, so:
1.1998 +// (2) A read cannot let unrelated NON-volatile memory refs that happen after
1.1999 +// the read float up to before the read. It's OK for non-volatile memory refs
1.2000 +// that happen before the volatile read to float down below it.
1.2001 +// (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
1.2002 +// that happen BEFORE the write float down to after the write. It's OK for
1.2003 +// non-volatile memory refs that happen after the volatile write to float up
1.2004 +// before it.
1.2005 +//
1.2006 +// We only put in barriers around volatile refs (they are expensive), not
1.2007 +// _between_ memory refs (that would require us to track the flavor of the
1.2008 +// previous memory refs). Requirements (2) and (3) require some barriers
1.2009 +// before volatile stores and after volatile loads. These nearly cover
1.2010 +// requirement (1) but miss the volatile-store-volatile-load case. This final
1.2011 +// case is placed after volatile-stores although it could just as well go
1.2012 +// before volatile-loads.
1.2013 +void TemplateTable::volatile_barrier( ) {
1.2014 + // Helper function to insert a is-volatile test and memory barrier
1.2015 + if( !os::is_MP() ) return; // Not needed on single CPU
1.2016 + __ membar();
1.2017 +}
1.2018 +
1.2019 +void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register index) {
1.2020 + assert(byte_no == 1 || byte_no == 2, "byte_no out of range");
1.2021 +
1.2022 + Register temp = rbx;
1.2023 +
1.2024 + assert_different_registers(Rcache, index, temp);
1.2025 +
1.2026 + const int shift_count = (1 + byte_no)*BitsPerByte;
1.2027 + Label resolved;
1.2028 + __ get_cache_and_index_at_bcp(Rcache, index, 1);
1.2029 + __ movl(temp, Address(Rcache, index, Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
1.2030 + __ shrl(temp, shift_count);
1.2031 + // have we resolved this bytecode?
1.2032 + __ andl(temp, 0xFF);
1.2033 + __ cmpl(temp, (int)bytecode());
1.2034 + __ jcc(Assembler::equal, resolved);
1.2035 +
1.2036 + // resolve first time through
1.2037 + address entry;
1.2038 + switch (bytecode()) {
1.2039 + case Bytecodes::_getstatic : // fall through
1.2040 + case Bytecodes::_putstatic : // fall through
1.2041 + case Bytecodes::_getfield : // fall through
1.2042 + case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
1.2043 + case Bytecodes::_invokevirtual : // fall through
1.2044 + case Bytecodes::_invokespecial : // fall through
1.2045 + case Bytecodes::_invokestatic : // fall through
1.2046 + case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
1.2047 + default : ShouldNotReachHere(); break;
1.2048 + }
1.2049 + __ movl(temp, (int)bytecode());
1.2050 + __ call_VM(noreg, entry, temp);
1.2051 + // Update registers with resolved info
1.2052 + __ get_cache_and_index_at_bcp(Rcache, index, 1);
1.2053 + __ bind(resolved);
1.2054 +}
1.2055 +
1.2056 +
1.2057 +// The cache and index registers must be set before call
1.2058 +void TemplateTable::load_field_cp_cache_entry(Register obj,
1.2059 + Register cache,
1.2060 + Register index,
1.2061 + Register off,
1.2062 + Register flags,
1.2063 + bool is_static = false) {
1.2064 + assert_different_registers(cache, index, flags, off);
1.2065 +
1.2066 + ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
1.2067 + // Field offset
1.2068 + __ movl(off, Address(cache, index, Address::times_4,
1.2069 + in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset())));
1.2070 + // Flags
1.2071 + __ movl(flags, Address(cache, index, Address::times_4,
1.2072 + in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset())));
1.2073 +
1.2074 + // klass overwrite register
1.2075 + if (is_static) {
1.2076 + __ movl(obj, Address(cache, index, Address::times_4,
1.2077 + in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset())));
1.2078 + }
1.2079 +}
1.2080 +
1.2081 +void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
1.2082 + Register method,
1.2083 + Register itable_index,
1.2084 + Register flags,
1.2085 + bool is_invokevirtual,
1.2086 + bool is_invokevfinal /*unused*/) {
1.2087 + // setup registers
1.2088 + const Register cache = rcx;
1.2089 + const Register index = rdx;
1.2090 + assert_different_registers(method, flags);
1.2091 + assert_different_registers(method, cache, index);
1.2092 + assert_different_registers(itable_index, flags);
1.2093 + assert_different_registers(itable_index, cache, index);
1.2094 + // determine constant pool cache field offsets
1.2095 + const int method_offset = in_bytes(
1.2096 + constantPoolCacheOopDesc::base_offset() +
1.2097 + (is_invokevirtual
1.2098 + ? ConstantPoolCacheEntry::f2_offset()
1.2099 + : ConstantPoolCacheEntry::f1_offset()
1.2100 + )
1.2101 + );
1.2102 + const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
1.2103 + ConstantPoolCacheEntry::flags_offset());
1.2104 + // access constant pool cache fields
1.2105 + const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
1.2106 + ConstantPoolCacheEntry::f2_offset());
1.2107 +
1.2108 + resolve_cache_and_index(byte_no, cache, index);
1.2109 +
1.2110 + assert(wordSize == 4, "adjust code below");
1.2111 + __ movl(method, Address(cache, index, Address::times_4, method_offset));
1.2112 + if (itable_index != noreg) {
1.2113 + __ movl(itable_index, Address(cache, index, Address::times_4, index_offset));
1.2114 + }
1.2115 + __ movl(flags , Address(cache, index, Address::times_4, flags_offset ));
1.2116 +}
1.2117 +
1.2118 +
1.2119 +// The registers cache and index expected to be set before call.
1.2120 +// Correct values of the cache and index registers are preserved.
1.2121 +void TemplateTable::jvmti_post_field_access(Register cache,
1.2122 + Register index,
1.2123 + bool is_static,
1.2124 + bool has_tos) {
1.2125 + if (JvmtiExport::can_post_field_access()) {
1.2126 + // Check to see if a field access watch has been set before we take
1.2127 + // the time to call into the VM.
1.2128 + Label L1;
1.2129 + assert_different_registers(cache, index, rax);
1.2130 + __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
1.2131 + __ testl(rax,rax);
1.2132 + __ jcc(Assembler::zero, L1);
1.2133 +
1.2134 + // cache entry pointer
1.2135 + __ addl(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
1.2136 + __ shll(index, LogBytesPerWord);
1.2137 + __ addl(cache, index);
1.2138 + if (is_static) {
1.2139 + __ movl(rax, 0); // NULL object reference
1.2140 + } else {
1.2141 + __ pop(atos); // Get the object
1.2142 + __ verify_oop(rax);
1.2143 + __ push(atos); // Restore stack state
1.2144 + }
1.2145 + // rax,: object pointer or NULL
1.2146 + // cache: cache entry pointer
1.2147 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
1.2148 + rax, cache);
1.2149 + __ get_cache_and_index_at_bcp(cache, index, 1);
1.2150 + __ bind(L1);
1.2151 + }
1.2152 +}
1.2153 +
1.2154 +void TemplateTable::pop_and_check_object(Register r) {
1.2155 + __ pop_ptr(r);
1.2156 + __ null_check(r); // for field access must check obj.
1.2157 + __ verify_oop(r);
1.2158 +}
1.2159 +
1.2160 +void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
1.2161 + transition(vtos, vtos);
1.2162 +
1.2163 + const Register cache = rcx;
1.2164 + const Register index = rdx;
1.2165 + const Register obj = rcx;
1.2166 + const Register off = rbx;
1.2167 + const Register flags = rax;
1.2168 +
1.2169 + resolve_cache_and_index(byte_no, cache, index);
1.2170 + jvmti_post_field_access(cache, index, is_static, false);
1.2171 + load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
1.2172 +
1.2173 + if (!is_static) pop_and_check_object(obj);
1.2174 +
1.2175 + const Address lo(obj, off, Address::times_1, 0*wordSize);
1.2176 + const Address hi(obj, off, Address::times_1, 1*wordSize);
1.2177 +
1.2178 + Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
1.2179 +
1.2180 + __ shrl(flags, ConstantPoolCacheEntry::tosBits);
1.2181 + assert(btos == 0, "change code, btos != 0");
1.2182 + // btos
1.2183 + __ andl(flags, 0x0f);
1.2184 + __ jcc(Assembler::notZero, notByte);
1.2185 +
1.2186 + __ load_signed_byte(rax, lo );
1.2187 + __ push(btos);
1.2188 + // Rewrite bytecode to be faster
1.2189 + if (!is_static) {
1.2190 + patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx);
1.2191 + }
1.2192 + __ jmp(Done);
1.2193 +
1.2194 + __ bind(notByte);
1.2195 + // itos
1.2196 + __ cmpl(flags, itos );
1.2197 + __ jcc(Assembler::notEqual, notInt);
1.2198 +
1.2199 + __ movl(rax, lo );
1.2200 + __ push(itos);
1.2201 + // Rewrite bytecode to be faster
1.2202 + if (!is_static) {
1.2203 + patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx);
1.2204 + }
1.2205 + __ jmp(Done);
1.2206 +
1.2207 + __ bind(notInt);
1.2208 + // atos
1.2209 + __ cmpl(flags, atos );
1.2210 + __ jcc(Assembler::notEqual, notObj);
1.2211 +
1.2212 + __ movl(rax, lo );
1.2213 + __ push(atos);
1.2214 + if (!is_static) {
1.2215 + patch_bytecode(Bytecodes::_fast_agetfield, rcx, rbx);
1.2216 + }
1.2217 + __ jmp(Done);
1.2218 +
1.2219 + __ bind(notObj);
1.2220 + // ctos
1.2221 + __ cmpl(flags, ctos );
1.2222 + __ jcc(Assembler::notEqual, notChar);
1.2223 +
1.2224 + __ load_unsigned_word(rax, lo );
1.2225 + __ push(ctos);
1.2226 + if (!is_static) {
1.2227 + patch_bytecode(Bytecodes::_fast_cgetfield, rcx, rbx);
1.2228 + }
1.2229 + __ jmp(Done);
1.2230 +
1.2231 + __ bind(notChar);
1.2232 + // stos
1.2233 + __ cmpl(flags, stos );
1.2234 + __ jcc(Assembler::notEqual, notShort);
1.2235 +
1.2236 + __ load_signed_word(rax, lo );
1.2237 + __ push(stos);
1.2238 + if (!is_static) {
1.2239 + patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx);
1.2240 + }
1.2241 + __ jmp(Done);
1.2242 +
1.2243 + __ bind(notShort);
1.2244 + // ltos
1.2245 + __ cmpl(flags, ltos );
1.2246 + __ jcc(Assembler::notEqual, notLong);
1.2247 +
1.2248 + // Generate code as if volatile. There just aren't enough registers to
1.2249 + // save that information and this code is faster than the test.
1.2250 + __ fild_d(lo); // Must load atomically
1.2251 + __ subl(rsp,2*wordSize); // Make space for store
1.2252 + __ fistp_d(Address(rsp,0));
1.2253 + __ popl(rax);
1.2254 + __ popl(rdx);
1.2255 +
1.2256 + __ push(ltos);
1.2257 + // Don't rewrite to _fast_lgetfield for potential volatile case.
1.2258 + __ jmp(Done);
1.2259 +
1.2260 + __ bind(notLong);
1.2261 + // ftos
1.2262 + __ cmpl(flags, ftos );
1.2263 + __ jcc(Assembler::notEqual, notFloat);
1.2264 +
1.2265 + __ fld_s(lo);
1.2266 + __ push(ftos);
1.2267 + if (!is_static) {
1.2268 + patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx);
1.2269 + }
1.2270 + __ jmp(Done);
1.2271 +
1.2272 + __ bind(notFloat);
1.2273 + // dtos
1.2274 + __ cmpl(flags, dtos );
1.2275 + __ jcc(Assembler::notEqual, notDouble);
1.2276 +
1.2277 + __ fld_d(lo);
1.2278 + __ push(dtos);
1.2279 + if (!is_static) {
1.2280 + patch_bytecode(Bytecodes::_fast_dgetfield, rcx, rbx);
1.2281 + }
1.2282 + __ jmpb(Done);
1.2283 +
1.2284 + __ bind(notDouble);
1.2285 +
1.2286 + __ stop("Bad state");
1.2287 +
1.2288 + __ bind(Done);
1.2289 + // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
1.2290 + // volatile_barrier( );
1.2291 +}
1.2292 +
1.2293 +
1.2294 +void TemplateTable::getfield(int byte_no) {
1.2295 + getfield_or_static(byte_no, false);
1.2296 +}
1.2297 +
1.2298 +
1.2299 +void TemplateTable::getstatic(int byte_no) {
1.2300 + getfield_or_static(byte_no, true);
1.2301 +}
1.2302 +
1.2303 +// The registers cache and index expected to be set before call.
1.2304 +// The function may destroy various registers, just not the cache and index registers.
1.2305 +void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
1.2306 +
1.2307 + ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
1.2308 +
1.2309 + if (JvmtiExport::can_post_field_modification()) {
1.2310 + // Check to see if a field modification watch has been set before we take
1.2311 + // the time to call into the VM.
1.2312 + Label L1;
1.2313 + assert_different_registers(cache, index, rax);
1.2314 + __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
1.2315 + __ testl(rax, rax);
1.2316 + __ jcc(Assembler::zero, L1);
1.2317 +
1.2318 + // The cache and index registers have been already set.
1.2319 + // This allows to eliminate this call but the cache and index
1.2320 + // registers have to be correspondingly used after this line.
1.2321 + __ get_cache_and_index_at_bcp(rax, rdx, 1);
1.2322 +
1.2323 + if (is_static) {
1.2324 + // Life is simple. Null out the object pointer.
1.2325 + __ xorl(rbx, rbx);
1.2326 + } else {
1.2327 + // Life is harder. The stack holds the value on top, followed by the object.
1.2328 + // We don't know the size of the value, though; it could be one or two words
1.2329 + // depending on its type. As a result, we must find the type to determine where
1.2330 + // the object is.
1.2331 + Label two_word, valsize_known;
1.2332 + __ movl(rcx, Address(rax, rdx, Address::times_4, in_bytes(cp_base_offset +
1.2333 + ConstantPoolCacheEntry::flags_offset())));
1.2334 + __ movl(rbx, rsp);
1.2335 + __ shrl(rcx, ConstantPoolCacheEntry::tosBits);
1.2336 + // Make sure we don't need to mask rcx for tosBits after the above shift
1.2337 + ConstantPoolCacheEntry::verify_tosBits();
1.2338 + __ cmpl(rcx, ltos);
1.2339 + __ jccb(Assembler::equal, two_word);
1.2340 + __ cmpl(rcx, dtos);
1.2341 + __ jccb(Assembler::equal, two_word);
1.2342 + __ addl(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
1.2343 + __ jmpb(valsize_known);
1.2344 +
1.2345 + __ bind(two_word);
1.2346 + __ addl(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
1.2347 +
1.2348 + __ bind(valsize_known);
1.2349 + // setup object pointer
1.2350 + __ movl(rbx, Address(rbx, 0));
1.2351 + }
1.2352 + // cache entry pointer
1.2353 + __ addl(rax, in_bytes(cp_base_offset));
1.2354 + __ shll(rdx, LogBytesPerWord);
1.2355 + __ addl(rax, rdx);
1.2356 + // object (tos)
1.2357 + __ movl(rcx, rsp);
1.2358 + // rbx,: object pointer set up above (NULL if static)
1.2359 + // rax,: cache entry pointer
1.2360 + // rcx: jvalue object on the stack
1.2361 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
1.2362 + rbx, rax, rcx);
1.2363 + __ get_cache_and_index_at_bcp(cache, index, 1);
1.2364 + __ bind(L1);
1.2365 + }
1.2366 +}
1.2367 +
1.2368 +
1.2369 +void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
1.2370 + transition(vtos, vtos);
1.2371 +
1.2372 + const Register cache = rcx;
1.2373 + const Register index = rdx;
1.2374 + const Register obj = rcx;
1.2375 + const Register off = rbx;
1.2376 + const Register flags = rax;
1.2377 +
1.2378 + resolve_cache_and_index(byte_no, cache, index);
1.2379 + jvmti_post_field_mod(cache, index, is_static);
1.2380 + load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
1.2381 +
1.2382 + // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
1.2383 + // volatile_barrier( );
1.2384 +
1.2385 + Label notVolatile, Done;
1.2386 + __ movl(rdx, flags);
1.2387 + __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
1.2388 + __ andl(rdx, 0x1);
1.2389 +
1.2390 + // field addresses
1.2391 + const Address lo(obj, off, Address::times_1, 0*wordSize);
1.2392 + const Address hi(obj, off, Address::times_1, 1*wordSize);
1.2393 +
1.2394 + Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
1.2395 +
1.2396 + __ shrl(flags, ConstantPoolCacheEntry::tosBits);
1.2397 + assert(btos == 0, "change code, btos != 0");
1.2398 + // btos
1.2399 + __ andl(flags, 0x0f);
1.2400 + __ jcc(Assembler::notZero, notByte);
1.2401 +
1.2402 + __ pop(btos);
1.2403 + if (!is_static) pop_and_check_object(obj);
1.2404 + __ movb(lo, rax );
1.2405 + if (!is_static) {
1.2406 + patch_bytecode(Bytecodes::_fast_bputfield, rcx, rbx);
1.2407 + }
1.2408 + __ jmp(Done);
1.2409 +
1.2410 + __ bind(notByte);
1.2411 + // itos
1.2412 + __ cmpl(flags, itos );
1.2413 + __ jcc(Assembler::notEqual, notInt);
1.2414 +
1.2415 + __ pop(itos);
1.2416 + if (!is_static) pop_and_check_object(obj);
1.2417 +
1.2418 + __ movl(lo, rax );
1.2419 + if (!is_static) {
1.2420 + patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx);
1.2421 + }
1.2422 + __ jmp(Done);
1.2423 +
1.2424 + __ bind(notInt);
1.2425 + // atos
1.2426 + __ cmpl(flags, atos );
1.2427 + __ jcc(Assembler::notEqual, notObj);
1.2428 +
1.2429 + __ pop(atos);
1.2430 + if (!is_static) pop_and_check_object(obj);
1.2431 +
1.2432 + __ movl(lo, rax );
1.2433 + __ store_check(obj, lo); // Need to mark card
1.2434 + if (!is_static) {
1.2435 + patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx);
1.2436 + }
1.2437 + __ jmp(Done);
1.2438 +
1.2439 + __ bind(notObj);
1.2440 + // ctos
1.2441 + __ cmpl(flags, ctos );
1.2442 + __ jcc(Assembler::notEqual, notChar);
1.2443 +
1.2444 + __ pop(ctos);
1.2445 + if (!is_static) pop_and_check_object(obj);
1.2446 + __ movw(lo, rax );
1.2447 + if (!is_static) {
1.2448 + patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx);
1.2449 + }
1.2450 + __ jmp(Done);
1.2451 +
1.2452 + __ bind(notChar);
1.2453 + // stos
1.2454 + __ cmpl(flags, stos );
1.2455 + __ jcc(Assembler::notEqual, notShort);
1.2456 +
1.2457 + __ pop(stos);
1.2458 + if (!is_static) pop_and_check_object(obj);
1.2459 + __ movw(lo, rax );
1.2460 + if (!is_static) {
1.2461 + patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx);
1.2462 + }
1.2463 + __ jmp(Done);
1.2464 +
1.2465 + __ bind(notShort);
1.2466 + // ltos
1.2467 + __ cmpl(flags, ltos );
1.2468 + __ jcc(Assembler::notEqual, notLong);
1.2469 +
1.2470 + Label notVolatileLong;
1.2471 + __ testl(rdx, rdx);
1.2472 + __ jcc(Assembler::zero, notVolatileLong);
1.2473 +
1.2474 + __ pop(ltos); // overwrites rdx, do this after testing volatile.
1.2475 + if (!is_static) pop_and_check_object(obj);
1.2476 +
1.2477 + // Replace with real volatile test
1.2478 + __ pushl(rdx);
1.2479 + __ pushl(rax); // Must update atomically with FIST
1.2480 + __ fild_d(Address(rsp,0)); // So load into FPU register
1.2481 + __ fistp_d(lo); // and put into memory atomically
1.2482 + __ addl(rsp,2*wordSize);
1.2483 + volatile_barrier();
1.2484 + // Don't rewrite volatile version
1.2485 + __ jmp(notVolatile);
1.2486 +
1.2487 + __ bind(notVolatileLong);
1.2488 +
1.2489 + __ pop(ltos); // overwrites rdx
1.2490 + if (!is_static) pop_and_check_object(obj);
1.2491 + __ movl(hi, rdx);
1.2492 + __ movl(lo, rax);
1.2493 + if (!is_static) {
1.2494 + patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx);
1.2495 + }
1.2496 + __ jmp(notVolatile);
1.2497 +
1.2498 + __ bind(notLong);
1.2499 + // ftos
1.2500 + __ cmpl(flags, ftos );
1.2501 + __ jcc(Assembler::notEqual, notFloat);
1.2502 +
1.2503 + __ pop(ftos);
1.2504 + if (!is_static) pop_and_check_object(obj);
1.2505 + __ fstp_s(lo);
1.2506 + if (!is_static) {
1.2507 + patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx);
1.2508 + }
1.2509 + __ jmp(Done);
1.2510 +
1.2511 + __ bind(notFloat);
1.2512 + // dtos
1.2513 + __ cmpl(flags, dtos );
1.2514 + __ jcc(Assembler::notEqual, notDouble);
1.2515 +
1.2516 + __ pop(dtos);
1.2517 + if (!is_static) pop_and_check_object(obj);
1.2518 + __ fstp_d(lo);
1.2519 + if (!is_static) {
1.2520 + patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx);
1.2521 + }
1.2522 + __ jmp(Done);
1.2523 +
1.2524 + __ bind(notDouble);
1.2525 +
1.2526 + __ stop("Bad state");
1.2527 +
1.2528 + __ bind(Done);
1.2529 +
1.2530 + // Check for volatile store
1.2531 + __ testl(rdx, rdx);
1.2532 + __ jcc(Assembler::zero, notVolatile);
1.2533 + volatile_barrier( );
1.2534 + __ bind(notVolatile);
1.2535 +}
1.2536 +
1.2537 +
1.2538 +void TemplateTable::putfield(int byte_no) {
1.2539 + putfield_or_static(byte_no, false);
1.2540 +}
1.2541 +
1.2542 +
1.2543 +void TemplateTable::putstatic(int byte_no) {
1.2544 + putfield_or_static(byte_no, true);
1.2545 +}
1.2546 +
1.2547 +void TemplateTable::jvmti_post_fast_field_mod() {
1.2548 + if (JvmtiExport::can_post_field_modification()) {
1.2549 + // Check to see if a field modification watch has been set before we take
1.2550 + // the time to call into the VM.
1.2551 + Label L2;
1.2552 + __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
1.2553 + __ testl(rcx,rcx);
1.2554 + __ jcc(Assembler::zero, L2);
1.2555 + __ pop_ptr(rbx); // copy the object pointer from tos
1.2556 + __ verify_oop(rbx);
1.2557 + __ push_ptr(rbx); // put the object pointer back on tos
1.2558 + __ subl(rsp, sizeof(jvalue)); // add space for a jvalue object
1.2559 + __ movl(rcx, rsp);
1.2560 + __ push_ptr(rbx); // save object pointer so we can steal rbx,
1.2561 + __ movl(rbx, 0);
1.2562 + const Address lo_value(rcx, rbx, Address::times_1, 0*wordSize);
1.2563 + const Address hi_value(rcx, rbx, Address::times_1, 1*wordSize);
1.2564 + switch (bytecode()) { // load values into the jvalue object
1.2565 + case Bytecodes::_fast_bputfield: __ movb(lo_value, rax); break;
1.2566 + case Bytecodes::_fast_sputfield: __ movw(lo_value, rax); break;
1.2567 + case Bytecodes::_fast_cputfield: __ movw(lo_value, rax); break;
1.2568 + case Bytecodes::_fast_iputfield: __ movl(lo_value, rax); break;
1.2569 + case Bytecodes::_fast_lputfield: __ movl(hi_value, rdx); __ movl(lo_value, rax); break;
1.2570 + // need to call fld_s() after fstp_s() to restore the value for below
1.2571 + case Bytecodes::_fast_fputfield: __ fstp_s(lo_value); __ fld_s(lo_value); break;
1.2572 + // need to call fld_d() after fstp_d() to restore the value for below
1.2573 + case Bytecodes::_fast_dputfield: __ fstp_d(lo_value); __ fld_d(lo_value); break;
1.2574 + // since rcx is not an object we don't call store_check() here
1.2575 + case Bytecodes::_fast_aputfield: __ movl(lo_value, rax); break;
1.2576 + default: ShouldNotReachHere();
1.2577 + }
1.2578 + __ pop_ptr(rbx); // restore copy of object pointer
1.2579 +
1.2580 + // Save rax, and sometimes rdx because call_VM() will clobber them,
1.2581 + // then use them for JVM/DI purposes
1.2582 + __ pushl(rax);
1.2583 + if (bytecode() == Bytecodes::_fast_lputfield) __ pushl(rdx);
1.2584 + // access constant pool cache entry
1.2585 + __ get_cache_entry_pointer_at_bcp(rax, rdx, 1);
1.2586 + __ verify_oop(rbx);
1.2587 + // rbx,: object pointer copied above
1.2588 + // rax,: cache entry pointer
1.2589 + // rcx: jvalue object on the stack
1.2590 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx);
1.2591 + if (bytecode() == Bytecodes::_fast_lputfield) __ popl(rdx); // restore high value
1.2592 + __ popl(rax); // restore lower value
1.2593 + __ addl(rsp, sizeof(jvalue)); // release jvalue object space
1.2594 + __ bind(L2);
1.2595 + }
1.2596 +}
1.2597 +
1.2598 +void TemplateTable::fast_storefield(TosState state) {
1.2599 + transition(state, vtos);
1.2600 +
1.2601 + ByteSize base = constantPoolCacheOopDesc::base_offset();
1.2602 +
1.2603 + jvmti_post_fast_field_mod();
1.2604 +
1.2605 + // access constant pool cache
1.2606 + __ get_cache_and_index_at_bcp(rcx, rbx, 1);
1.2607 +
1.2608 + // test for volatile with rdx but rdx is tos register for lputfield.
1.2609 + if (bytecode() == Bytecodes::_fast_lputfield) __ pushl(rdx);
1.2610 + __ movl(rdx, Address(rcx, rbx, Address::times_4, in_bytes(base +
1.2611 + ConstantPoolCacheEntry::flags_offset())));
1.2612 +
1.2613 + // replace index with field offset from cache entry
1.2614 + __ movl(rbx, Address(rcx, rbx, Address::times_4, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
1.2615 +
1.2616 + // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
1.2617 + // volatile_barrier( );
1.2618 +
1.2619 + Label notVolatile, Done;
1.2620 + __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
1.2621 + __ andl(rdx, 0x1);
1.2622 + // Check for volatile store
1.2623 + __ testl(rdx, rdx);
1.2624 + __ jcc(Assembler::zero, notVolatile);
1.2625 +
1.2626 + if (bytecode() == Bytecodes::_fast_lputfield) __ popl(rdx);
1.2627 +
1.2628 + // Get object from stack
1.2629 + pop_and_check_object(rcx);
1.2630 +
1.2631 + // field addresses
1.2632 + const Address lo(rcx, rbx, Address::times_1, 0*wordSize);
1.2633 + const Address hi(rcx, rbx, Address::times_1, 1*wordSize);
1.2634 +
1.2635 + // access field
1.2636 + switch (bytecode()) {
1.2637 + case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
1.2638 + case Bytecodes::_fast_sputfield: // fall through
1.2639 + case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
1.2640 + case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
1.2641 + case Bytecodes::_fast_lputfield: __ movl(hi, rdx); __ movl(lo, rax); break;
1.2642 + case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
1.2643 + case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
1.2644 + case Bytecodes::_fast_aputfield: __ movl(lo, rax); __ store_check(rcx, lo); break;
1.2645 + default:
1.2646 + ShouldNotReachHere();
1.2647 + }
1.2648 +
1.2649 + Label done;
1.2650 + volatile_barrier( );
1.2651 + __ jmpb(done);
1.2652 +
1.2653 + // Same code as above, but don't need rdx to test for volatile.
1.2654 + __ bind(notVolatile);
1.2655 +
1.2656 + if (bytecode() == Bytecodes::_fast_lputfield) __ popl(rdx);
1.2657 +
1.2658 + // Get object from stack
1.2659 + pop_and_check_object(rcx);
1.2660 +
1.2661 + // access field
1.2662 + switch (bytecode()) {
1.2663 + case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
1.2664 + case Bytecodes::_fast_sputfield: // fall through
1.2665 + case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
1.2666 + case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
1.2667 + case Bytecodes::_fast_lputfield: __ movl(hi, rdx); __ movl(lo, rax); break;
1.2668 + case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
1.2669 + case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
1.2670 + case Bytecodes::_fast_aputfield: __ movl(lo, rax); __ store_check(rcx, lo); break;
1.2671 + default:
1.2672 + ShouldNotReachHere();
1.2673 + }
1.2674 + __ bind(done);
1.2675 +}
1.2676 +
1.2677 +
1.2678 +void TemplateTable::fast_accessfield(TosState state) {
1.2679 + transition(atos, state);
1.2680 +
1.2681 + // do the JVMTI work here to avoid disturbing the register state below
1.2682 + if (JvmtiExport::can_post_field_access()) {
1.2683 + // Check to see if a field access watch has been set before we take
1.2684 + // the time to call into the VM.
1.2685 + Label L1;
1.2686 + __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
1.2687 + __ testl(rcx,rcx);
1.2688 + __ jcc(Assembler::zero, L1);
1.2689 + // access constant pool cache entry
1.2690 + __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1);
1.2691 + __ push_ptr(rax); // save object pointer before call_VM() clobbers it
1.2692 + __ verify_oop(rax);
1.2693 + // rax,: object pointer copied above
1.2694 + // rcx: cache entry pointer
1.2695 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx);
1.2696 + __ pop_ptr(rax); // restore object pointer
1.2697 + __ bind(L1);
1.2698 + }
1.2699 +
1.2700 + // access constant pool cache
1.2701 + __ get_cache_and_index_at_bcp(rcx, rbx, 1);
1.2702 + // replace index with field offset from cache entry
1.2703 + __ movl(rbx, Address(rcx, rbx, Address::times_4, in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
1.2704 +
1.2705 +
1.2706 + // rax,: object
1.2707 + __ verify_oop(rax);
1.2708 + __ null_check(rax);
1.2709 + // field addresses
1.2710 + const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
1.2711 + const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize);
1.2712 +
1.2713 + // access field
1.2714 + switch (bytecode()) {
1.2715 + case Bytecodes::_fast_bgetfield: __ movsxb(rax, lo ); break;
1.2716 + case Bytecodes::_fast_sgetfield: __ load_signed_word(rax, lo ); break;
1.2717 + case Bytecodes::_fast_cgetfield: __ load_unsigned_word(rax, lo ); break;
1.2718 + case Bytecodes::_fast_igetfield: __ movl(rax, lo); break;
1.2719 + case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten"); break;
1.2720 + case Bytecodes::_fast_fgetfield: __ fld_s(lo); break;
1.2721 + case Bytecodes::_fast_dgetfield: __ fld_d(lo); break;
1.2722 + case Bytecodes::_fast_agetfield: __ movl(rax, lo); __ verify_oop(rax); break;
1.2723 + default:
1.2724 + ShouldNotReachHere();
1.2725 + }
1.2726 +
1.2727 + // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO)
1.2728 + // volatile_barrier( );
1.2729 +}
1.2730 +
1.2731 +void TemplateTable::fast_xaccess(TosState state) {
1.2732 + transition(vtos, state);
1.2733 + // get receiver
1.2734 + __ movl(rax, aaddress(0));
1.2735 + debug_only(__ verify_local_tag(frame::TagReference, 0));
1.2736 + // access constant pool cache
1.2737 + __ get_cache_and_index_at_bcp(rcx, rdx, 2);
1.2738 + __ movl(rbx, Address(rcx, rdx, Address::times_4, in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
1.2739 + // make sure exception is reported in correct bcp range (getfield is next instruction)
1.2740 + __ increment(rsi);
1.2741 + __ null_check(rax);
1.2742 + const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
1.2743 + if (state == itos) {
1.2744 + __ movl(rax, lo);
1.2745 + } else if (state == atos) {
1.2746 + __ movl(rax, lo);
1.2747 + __ verify_oop(rax);
1.2748 + } else if (state == ftos) {
1.2749 + __ fld_s(lo);
1.2750 + } else {
1.2751 + ShouldNotReachHere();
1.2752 + }
1.2753 + __ decrement(rsi);
1.2754 +}
1.2755 +
1.2756 +
1.2757 +
1.2758 +//----------------------------------------------------------------------------------------------------
1.2759 +// Calls
1.2760 +
1.2761 +void TemplateTable::count_calls(Register method, Register temp) {
1.2762 + // implemented elsewhere
1.2763 + ShouldNotReachHere();
1.2764 +}
1.2765 +
1.2766 +
1.2767 +void TemplateTable::prepare_invoke(Register method, Register index, int byte_no, Bytecodes::Code code) {
1.2768 + // determine flags
1.2769 + const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
1.2770 + const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
1.2771 + const bool is_invokespecial = code == Bytecodes::_invokespecial;
1.2772 + const bool load_receiver = code != Bytecodes::_invokestatic;
1.2773 + const bool receiver_null_check = is_invokespecial;
1.2774 + const bool save_flags = is_invokeinterface || is_invokevirtual;
1.2775 + // setup registers & access constant pool cache
1.2776 + const Register recv = rcx;
1.2777 + const Register flags = rdx;
1.2778 + assert_different_registers(method, index, recv, flags);
1.2779 +
1.2780 + // save 'interpreter return address'
1.2781 + __ save_bcp();
1.2782 +
1.2783 + load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual);
1.2784 +
1.2785 + // load receiver if needed (note: no return address pushed yet)
1.2786 + if (load_receiver) {
1.2787 + __ movl(recv, flags);
1.2788 + __ andl(recv, 0xFF);
1.2789 + // recv count is 0 based?
1.2790 + __ movl(recv, Address(rsp, recv, Interpreter::stackElementScale(), -Interpreter::expr_offset_in_bytes(1)));
1.2791 + __ verify_oop(recv);
1.2792 + }
1.2793 +
1.2794 + // do null check if needed
1.2795 + if (receiver_null_check) {
1.2796 + __ null_check(recv);
1.2797 + }
1.2798 +
1.2799 + if (save_flags) {
1.2800 + __ movl(rsi, flags);
1.2801 + }
1.2802 +
1.2803 + // compute return type
1.2804 + __ shrl(flags, ConstantPoolCacheEntry::tosBits);
1.2805 + // Make sure we don't need to mask flags for tosBits after the above shift
1.2806 + ConstantPoolCacheEntry::verify_tosBits();
1.2807 + // load return address
1.2808 + { const int table =
1.2809 + is_invokeinterface
1.2810 + ? (int)Interpreter::return_5_addrs_by_index_table()
1.2811 + : (int)Interpreter::return_3_addrs_by_index_table();
1.2812 + __ movl(flags, Address(noreg, flags, Address::times_4, table));
1.2813 + }
1.2814 +
1.2815 + // push return address
1.2816 + __ pushl(flags);
1.2817 +
1.2818 + // Restore flag value from the constant pool cache, and restore rsi
1.2819 + // for later null checks. rsi is the bytecode pointer
1.2820 + if (save_flags) {
1.2821 + __ movl(flags, rsi);
1.2822 + __ restore_bcp();
1.2823 + }
1.2824 +}
1.2825 +
1.2826 +
1.2827 +void TemplateTable::invokevirtual_helper(Register index, Register recv,
1.2828 + Register flags) {
1.2829 +
1.2830 + // Uses temporary registers rax, rdx
1.2831 + assert_different_registers(index, recv, rax, rdx);
1.2832 +
1.2833 + // Test for an invoke of a final method
1.2834 + Label notFinal;
1.2835 + __ movl(rax, flags);
1.2836 + __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod));
1.2837 + __ jcc(Assembler::zero, notFinal);
1.2838 +
1.2839 + Register method = index; // method must be rbx,
1.2840 + assert(method == rbx, "methodOop must be rbx, for interpreter calling convention");
1.2841 +
1.2842 + // do the call - the index is actually the method to call
1.2843 + __ verify_oop(method);
1.2844 +
1.2845 + // It's final, need a null check here!
1.2846 + __ null_check(recv);
1.2847 +
1.2848 + // profile this call
1.2849 + __ profile_final_call(rax);
1.2850 +
1.2851 + __ jump_from_interpreted(method, rax);
1.2852 +
1.2853 + __ bind(notFinal);
1.2854 +
1.2855 + // get receiver klass
1.2856 + __ null_check(recv, oopDesc::klass_offset_in_bytes());
1.2857 + // Keep recv in rcx for callee expects it there
1.2858 + __ movl(rax, Address(recv, oopDesc::klass_offset_in_bytes()));
1.2859 + __ verify_oop(rax);
1.2860 +
1.2861 + // profile this call
1.2862 + __ profile_virtual_call(rax, rdi, rdx);
1.2863 +
1.2864 + // get target methodOop & entry point
1.2865 + const int base = instanceKlass::vtable_start_offset() * wordSize;
1.2866 + assert(vtableEntry::size() * wordSize == 4, "adjust the scaling in the code below");
1.2867 + __ movl(method, Address(rax, index, Address::times_4, base + vtableEntry::method_offset_in_bytes()));
1.2868 + __ jump_from_interpreted(method, rdx);
1.2869 +}
1.2870 +
1.2871 +
1.2872 +void TemplateTable::invokevirtual(int byte_no) {
1.2873 + transition(vtos, vtos);
1.2874 + prepare_invoke(rbx, noreg, byte_no, bytecode());
1.2875 +
1.2876 + // rbx,: index
1.2877 + // rcx: receiver
1.2878 + // rdx: flags
1.2879 +
1.2880 + invokevirtual_helper(rbx, rcx, rdx);
1.2881 +}
1.2882 +
1.2883 +
1.2884 +void TemplateTable::invokespecial(int byte_no) {
1.2885 + transition(vtos, vtos);
1.2886 + prepare_invoke(rbx, noreg, byte_no, bytecode());
1.2887 + // do the call
1.2888 + __ verify_oop(rbx);
1.2889 + __ profile_call(rax);
1.2890 + __ jump_from_interpreted(rbx, rax);
1.2891 +}
1.2892 +
1.2893 +
1.2894 +void TemplateTable::invokestatic(int byte_no) {
1.2895 + transition(vtos, vtos);
1.2896 + prepare_invoke(rbx, noreg, byte_no, bytecode());
1.2897 + // do the call
1.2898 + __ verify_oop(rbx);
1.2899 + __ profile_call(rax);
1.2900 + __ jump_from_interpreted(rbx, rax);
1.2901 +}
1.2902 +
1.2903 +
1.2904 +void TemplateTable::fast_invokevfinal(int byte_no) {
1.2905 + transition(vtos, vtos);
1.2906 + __ stop("fast_invokevfinal not used on x86");
1.2907 +}
1.2908 +
1.2909 +
1.2910 +void TemplateTable::invokeinterface(int byte_no) {
1.2911 + transition(vtos, vtos);
1.2912 + prepare_invoke(rax, rbx, byte_no, bytecode());
1.2913 +
1.2914 + // rax,: Interface
1.2915 + // rbx,: index
1.2916 + // rcx: receiver
1.2917 + // rdx: flags
1.2918 +
1.2919 + // Special case of invokeinterface called for virtual method of
1.2920 + // java.lang.Object. See cpCacheOop.cpp for details.
1.2921 + // This code isn't produced by javac, but could be produced by
1.2922 + // another compliant java compiler.
1.2923 + Label notMethod;
1.2924 + __ movl(rdi, rdx);
1.2925 + __ andl(rdi, (1 << ConstantPoolCacheEntry::methodInterface));
1.2926 + __ jcc(Assembler::zero, notMethod);
1.2927 +
1.2928 + invokevirtual_helper(rbx, rcx, rdx);
1.2929 + __ bind(notMethod);
1.2930 +
1.2931 + // Get receiver klass into rdx - also a null check
1.2932 + __ restore_locals(); // restore rdi
1.2933 + __ movl(rdx, Address(rcx, oopDesc::klass_offset_in_bytes()));
1.2934 + __ verify_oop(rdx);
1.2935 +
1.2936 + // profile this call
1.2937 + __ profile_virtual_call(rdx, rsi, rdi);
1.2938 +
1.2939 + __ movl(rdi, rdx); // Save klassOop in rdi
1.2940 +
1.2941 + // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1.2942 + const int base = instanceKlass::vtable_start_offset() * wordSize;
1.2943 + assert(vtableEntry::size() * wordSize == 4, "adjust the scaling in the code below");
1.2944 + __ movl(rsi, Address(rdx, instanceKlass::vtable_length_offset() * wordSize)); // Get length of vtable
1.2945 + __ leal(rdx, Address(rdx, rsi, Address::times_4, base));
1.2946 + if (HeapWordsPerLong > 1) {
1.2947 + // Round up to align_object_offset boundary
1.2948 + __ round_to(rdx, BytesPerLong);
1.2949 + }
1.2950 +
1.2951 + Label entry, search, interface_ok;
1.2952 +
1.2953 + __ jmpb(entry);
1.2954 + __ bind(search);
1.2955 + __ addl(rdx, itableOffsetEntry::size() * wordSize);
1.2956 +
1.2957 + __ bind(entry);
1.2958 +
1.2959 + // Check that the entry is non-null. A null entry means that the receiver
1.2960 + // class doesn't implement the interface, and wasn't the same as the
1.2961 + // receiver class checked when the interface was resolved.
1.2962 + __ pushl(rdx);
1.2963 + __ movl(rdx, Address(rdx, itableOffsetEntry::interface_offset_in_bytes()));
1.2964 + __ testl(rdx, rdx);
1.2965 + __ jcc(Assembler::notZero, interface_ok);
1.2966 + // throw exception
1.2967 + __ popl(rdx); // pop saved register first.
1.2968 + __ popl(rbx); // pop return address (pushed by prepare_invoke)
1.2969 + __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1.2970 + __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1.2971 + __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1.2972 + InterpreterRuntime::throw_IncompatibleClassChangeError));
1.2973 + // the call_VM checks for exception, so we should never return here.
1.2974 + __ should_not_reach_here();
1.2975 + __ bind(interface_ok);
1.2976 +
1.2977 + __ popl(rdx);
1.2978 +
1.2979 + __ cmpl(rax, Address(rdx, itableOffsetEntry::interface_offset_in_bytes()));
1.2980 + __ jcc(Assembler::notEqual, search);
1.2981 +
1.2982 + __ movl(rdx, Address(rdx, itableOffsetEntry::offset_offset_in_bytes()));
1.2983 + __ addl(rdx, rdi); // Add offset to klassOop
1.2984 + assert(itableMethodEntry::size() * wordSize == 4, "adjust the scaling in the code below");
1.2985 + __ movl(rbx, Address(rdx, rbx, Address::times_4));
1.2986 + // rbx,: methodOop to call
1.2987 + // rcx: receiver
1.2988 + // Check for abstract method error
1.2989 + // Note: This should be done more efficiently via a throw_abstract_method_error
1.2990 + // interpreter entry point and a conditional jump to it in case of a null
1.2991 + // method.
1.2992 + { Label L;
1.2993 + __ testl(rbx, rbx);
1.2994 + __ jcc(Assembler::notZero, L);
1.2995 + // throw exception
1.2996 + // note: must restore interpreter registers to canonical
1.2997 + // state for exception handling to work correctly!
1.2998 + __ popl(rbx); // pop return address (pushed by prepare_invoke)
1.2999 + __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
1.3000 + __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
1.3001 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
1.3002 + // the call_VM checks for exception, so we should never return here.
1.3003 + __ should_not_reach_here();
1.3004 + __ bind(L);
1.3005 + }
1.3006 +
1.3007 + // do the call
1.3008 + // rcx: receiver
1.3009 + // rbx,: methodOop
1.3010 + __ jump_from_interpreted(rbx, rdx);
1.3011 +}
1.3012 +
1.3013 +//----------------------------------------------------------------------------------------------------
1.3014 +// Allocation
1.3015 +
1.3016 +void TemplateTable::_new() {
1.3017 + transition(vtos, atos);
1.3018 + __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
1.3019 + Label slow_case;
1.3020 + Label done;
1.3021 + Label initialize_header;
1.3022 + Label initialize_object; // including clearing the fields
1.3023 + Label allocate_shared;
1.3024 +
1.3025 + ExternalAddress heap_top((address)Universe::heap()->top_addr());
1.3026 +
1.3027 + __ get_cpool_and_tags(rcx, rax);
1.3028 + // get instanceKlass
1.3029 + __ movl(rcx, Address(rcx, rdx, Address::times_4, sizeof(constantPoolOopDesc)));
1.3030 + __ pushl(rcx); // save the contexts of klass for initializing the header
1.3031 +
1.3032 + // make sure the class we're about to instantiate has been resolved.
1.3033 + // Note: slow_case does a pop of stack, which is why we loaded class/pushed above
1.3034 + const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
1.3035 + __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
1.3036 + __ jcc(Assembler::notEqual, slow_case);
1.3037 +
1.3038 + // make sure klass is initialized & doesn't have finalizer
1.3039 + // make sure klass is fully initialized
1.3040 + __ cmpl(Address(rcx, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)), instanceKlass::fully_initialized);
1.3041 + __ jcc(Assembler::notEqual, slow_case);
1.3042 +
1.3043 + // get instance_size in instanceKlass (scaled to a count of bytes)
1.3044 + __ movl(rdx, Address(rcx, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
1.3045 + // test to see if it has a finalizer or is malformed in some way
1.3046 + __ testl(rdx, Klass::_lh_instance_slow_path_bit);
1.3047 + __ jcc(Assembler::notZero, slow_case);
1.3048 +
1.3049 + //
1.3050 + // Allocate the instance
1.3051 + // 1) Try to allocate in the TLAB
1.3052 + // 2) if fail and the object is large allocate in the shared Eden
1.3053 + // 3) if the above fails (or is not applicable), go to a slow case
1.3054 + // (creates a new TLAB, etc.)
1.3055 +
1.3056 + const bool allow_shared_alloc =
1.3057 + Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
1.3058 +
1.3059 + if (UseTLAB) {
1.3060 + const Register thread = rcx;
1.3061 +
1.3062 + __ get_thread(thread);
1.3063 + __ movl(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
1.3064 + __ leal(rbx, Address(rax, rdx, Address::times_1));
1.3065 + __ cmpl(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
1.3066 + __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
1.3067 + __ movl(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
1.3068 + if (ZeroTLAB) {
1.3069 + // the fields have been already cleared
1.3070 + __ jmp(initialize_header);
1.3071 + } else {
1.3072 + // initialize both the header and fields
1.3073 + __ jmp(initialize_object);
1.3074 + }
1.3075 + }
1.3076 +
1.3077 + // Allocation in the shared Eden, if allowed.
1.3078 + //
1.3079 + // rdx: instance size in bytes
1.3080 + if (allow_shared_alloc) {
1.3081 + __ bind(allocate_shared);
1.3082 +
1.3083 + Label retry;
1.3084 + __ bind(retry);
1.3085 + __ mov32(rax, heap_top);
1.3086 + __ leal(rbx, Address(rax, rdx, Address::times_1));
1.3087 + __ cmp32(rbx, ExternalAddress((address)Universe::heap()->end_addr()));
1.3088 + __ jcc(Assembler::above, slow_case);
1.3089 +
1.3090 + // Compare rax, with the top addr, and if still equal, store the new
1.3091 + // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
1.3092 + // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
1.3093 + //
1.3094 + // rax,: object begin
1.3095 + // rbx,: object end
1.3096 + // rdx: instance size in bytes
1.3097 + if (os::is_MP()) __ lock();
1.3098 + __ cmpxchgptr(rbx, heap_top);
1.3099 +
1.3100 + // if someone beat us on the allocation, try again, otherwise continue
1.3101 + __ jcc(Assembler::notEqual, retry);
1.3102 + }
1.3103 +
1.3104 + if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
1.3105 + // The object is initialized before the header. If the object size is
1.3106 + // zero, go directly to the header initialization.
1.3107 + __ bind(initialize_object);
1.3108 + __ decrement(rdx, sizeof(oopDesc));
1.3109 + __ jcc(Assembler::zero, initialize_header);
1.3110 +
1.3111 + // Initialize topmost object field, divide rdx by 8, check if odd and
1.3112 + // test if zero.
1.3113 + __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
1.3114 + __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
1.3115 +
1.3116 + // rdx must have been multiple of 8
1.3117 +#ifdef ASSERT
1.3118 + // make sure rdx was multiple of 8
1.3119 + Label L;
1.3120 + // Ignore partial flag stall after shrl() since it is debug VM
1.3121 + __ jccb(Assembler::carryClear, L);
1.3122 + __ stop("object size is not multiple of 2 - adjust this code");
1.3123 + __ bind(L);
1.3124 + // rdx must be > 0, no extra check needed here
1.3125 +#endif
1.3126 +
1.3127 + // initialize remaining object fields: rdx was a multiple of 8
1.3128 + { Label loop;
1.3129 + __ bind(loop);
1.3130 + __ movl(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
1.3131 + __ movl(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx);
1.3132 + __ decrement(rdx);
1.3133 + __ jcc(Assembler::notZero, loop);
1.3134 + }
1.3135 +
1.3136 + // initialize object header only.
1.3137 + __ bind(initialize_header);
1.3138 + if (UseBiasedLocking) {
1.3139 + __ popl(rcx); // get saved klass back in the register.
1.3140 + __ movl(rbx, Address(rcx, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
1.3141 + __ movl(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
1.3142 + } else {
1.3143 + __ movl(Address(rax, oopDesc::mark_offset_in_bytes ()),
1.3144 + (int)markOopDesc::prototype()); // header
1.3145 + __ popl(rcx); // get saved klass back in the register.
1.3146 + }
1.3147 + __ movl(Address(rax, oopDesc::klass_offset_in_bytes()), rcx); // klass
1.3148 +
1.3149 + {
1.3150 + SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
1.3151 + // Trigger dtrace event for fastpath
1.3152 + __ push(atos);
1.3153 + __ call_VM_leaf(
1.3154 + CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
1.3155 + __ pop(atos);
1.3156 + }
1.3157 +
1.3158 + __ jmp(done);
1.3159 + }
1.3160 +
1.3161 + // slow case
1.3162 + __ bind(slow_case);
1.3163 + __ popl(rcx); // restore stack pointer to what it was when we came in.
1.3164 + __ get_constant_pool(rax);
1.3165 + __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
1.3166 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx);
1.3167 +
1.3168 + // continue
1.3169 + __ bind(done);
1.3170 +}
1.3171 +
1.3172 +
1.3173 +void TemplateTable::newarray() {
1.3174 + transition(itos, atos);
1.3175 + __ push_i(rax); // make sure everything is on the stack
1.3176 + __ load_unsigned_byte(rdx, at_bcp(1));
1.3177 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax);
1.3178 + __ pop_i(rdx); // discard size
1.3179 +}
1.3180 +
1.3181 +
1.3182 +void TemplateTable::anewarray() {
1.3183 + transition(itos, atos);
1.3184 + __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
1.3185 + __ get_constant_pool(rcx);
1.3186 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax);
1.3187 +}
1.3188 +
1.3189 +
1.3190 +void TemplateTable::arraylength() {
1.3191 + transition(atos, itos);
1.3192 + __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
1.3193 + __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
1.3194 +}
1.3195 +
1.3196 +
1.3197 +void TemplateTable::checkcast() {
1.3198 + transition(atos, atos);
1.3199 + Label done, is_null, ok_is_subtype, quicked, resolved;
1.3200 + __ testl(rax, rax); // Object is in EAX
1.3201 + __ jcc(Assembler::zero, is_null);
1.3202 +
1.3203 + // Get cpool & tags index
1.3204 + __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
1.3205 + __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
1.3206 + // See if bytecode has already been quicked
1.3207 + __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
1.3208 + __ jcc(Assembler::equal, quicked);
1.3209 +
1.3210 + __ push(atos);
1.3211 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
1.3212 + __ pop_ptr(rdx);
1.3213 + __ jmpb(resolved);
1.3214 +
1.3215 + // Get superklass in EAX and subklass in EBX
1.3216 + __ bind(quicked);
1.3217 + __ movl(rdx, rax); // Save object in EDX; EAX needed for subtype check
1.3218 + __ movl(rax, Address(rcx, rbx, Address::times_4, sizeof(constantPoolOopDesc)));
1.3219 +
1.3220 + __ bind(resolved);
1.3221 + __ movl(rbx, Address(rdx, oopDesc::klass_offset_in_bytes()));
1.3222 +
1.3223 + // Generate subtype check. Blows ECX. Resets EDI. Object in EDX.
1.3224 + // Superklass in EAX. Subklass in EBX.
1.3225 + __ gen_subtype_check( rbx, ok_is_subtype );
1.3226 +
1.3227 + // Come here on failure
1.3228 + __ pushl(rdx);
1.3229 + // object is at TOS
1.3230 + __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
1.3231 +
1.3232 + // Come here on success
1.3233 + __ bind(ok_is_subtype);
1.3234 + __ movl(rax,rdx); // Restore object in EDX
1.3235 +
1.3236 + // Collect counts on whether this check-cast sees NULLs a lot or not.
1.3237 + if (ProfileInterpreter) {
1.3238 + __ jmp(done);
1.3239 + __ bind(is_null);
1.3240 + __ profile_null_seen(rcx);
1.3241 + } else {
1.3242 + __ bind(is_null); // same as 'done'
1.3243 + }
1.3244 + __ bind(done);
1.3245 +}
1.3246 +
1.3247 +
1.3248 +void TemplateTable::instanceof() {
1.3249 + transition(atos, itos);
1.3250 + Label done, is_null, ok_is_subtype, quicked, resolved;
1.3251 + __ testl(rax, rax);
1.3252 + __ jcc(Assembler::zero, is_null);
1.3253 +
1.3254 + // Get cpool & tags index
1.3255 + __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
1.3256 + __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
1.3257 + // See if bytecode has already been quicked
1.3258 + __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
1.3259 + __ jcc(Assembler::equal, quicked);
1.3260 +
1.3261 + __ push(atos);
1.3262 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
1.3263 + __ pop_ptr(rdx);
1.3264 + __ movl(rdx, Address(rdx, oopDesc::klass_offset_in_bytes()));
1.3265 + __ jmp(resolved);
1.3266 +
1.3267 + // Get superklass in EAX and subklass in EDX
1.3268 + __ bind(quicked);
1.3269 + __ movl(rdx, Address(rax, oopDesc::klass_offset_in_bytes()));
1.3270 + __ movl(rax, Address(rcx, rbx, Address::times_4, sizeof(constantPoolOopDesc)));
1.3271 +
1.3272 + __ bind(resolved);
1.3273 +
1.3274 + // Generate subtype check. Blows ECX. Resets EDI.
1.3275 + // Superklass in EAX. Subklass in EDX.
1.3276 + __ gen_subtype_check( rdx, ok_is_subtype );
1.3277 +
1.3278 + // Come here on failure
1.3279 + __ xorl(rax,rax);
1.3280 + __ jmpb(done);
1.3281 + // Come here on success
1.3282 + __ bind(ok_is_subtype);
1.3283 + __ movl(rax, 1);
1.3284 +
1.3285 + // Collect counts on whether this test sees NULLs a lot or not.
1.3286 + if (ProfileInterpreter) {
1.3287 + __ jmp(done);
1.3288 + __ bind(is_null);
1.3289 + __ profile_null_seen(rcx);
1.3290 + } else {
1.3291 + __ bind(is_null); // same as 'done'
1.3292 + }
1.3293 + __ bind(done);
1.3294 + // rax, = 0: obj == NULL or obj is not an instanceof the specified klass
1.3295 + // rax, = 1: obj != NULL and obj is an instanceof the specified klass
1.3296 +}
1.3297 +
1.3298 +
1.3299 +//----------------------------------------------------------------------------------------------------
1.3300 +// Breakpoints
1.3301 +void TemplateTable::_breakpoint() {
1.3302 +
1.3303 + // Note: We get here even if we are single stepping..
1.3304 + // jbug inists on setting breakpoints at every bytecode
1.3305 + // even if we are in single step mode.
1.3306 +
1.3307 + transition(vtos, vtos);
1.3308 +
1.3309 + // get the unpatched byte code
1.3310 + __ get_method(rcx);
1.3311 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi);
1.3312 + __ movl(rbx, rax);
1.3313 +
1.3314 + // post the breakpoint event
1.3315 + __ get_method(rcx);
1.3316 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi);
1.3317 +
1.3318 + // complete the execution of original bytecode
1.3319 + __ dispatch_only_normal(vtos);
1.3320 +}
1.3321 +
1.3322 +
1.3323 +//----------------------------------------------------------------------------------------------------
1.3324 +// Exceptions
1.3325 +
1.3326 +void TemplateTable::athrow() {
1.3327 + transition(atos, vtos);
1.3328 + __ null_check(rax);
1.3329 + __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
1.3330 +}
1.3331 +
1.3332 +
1.3333 +//----------------------------------------------------------------------------------------------------
1.3334 +// Synchronization
1.3335 +//
1.3336 +// Note: monitorenter & exit are symmetric routines; which is reflected
1.3337 +// in the assembly code structure as well
1.3338 +//
1.3339 +// Stack layout:
1.3340 +//
1.3341 +// [expressions ] <--- rsp = expression stack top
1.3342 +// ..
1.3343 +// [expressions ]
1.3344 +// [monitor entry] <--- monitor block top = expression stack bot
1.3345 +// ..
1.3346 +// [monitor entry]
1.3347 +// [frame data ] <--- monitor block bot
1.3348 +// ...
1.3349 +// [saved rbp, ] <--- rbp,
1.3350 +
1.3351 +
1.3352 +void TemplateTable::monitorenter() {
1.3353 + transition(atos, vtos);
1.3354 +
1.3355 + // check for NULL object
1.3356 + __ null_check(rax);
1.3357 +
1.3358 + const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1.3359 + const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1.3360 + const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
1.3361 + Label allocated;
1.3362 +
1.3363 + // initialize entry pointer
1.3364 + __ xorl(rdx, rdx); // points to free slot or NULL
1.3365 +
1.3366 + // find a free slot in the monitor block (result in rdx)
1.3367 + { Label entry, loop, exit;
1.3368 + __ movl(rcx, monitor_block_top); // points to current entry, starting with top-most entry
1.3369 + __ leal(rbx, monitor_block_bot); // points to word before bottom of monitor block
1.3370 + __ jmpb(entry);
1.3371 +
1.3372 + __ bind(loop);
1.3373 + __ cmpl(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), NULL_WORD); // check if current entry is used
1.3374 +
1.3375 +// TODO - need new func here - kbt
1.3376 + if (VM_Version::supports_cmov()) {
1.3377 + __ cmovl(Assembler::equal, rdx, rcx); // if not used then remember entry in rdx
1.3378 + } else {
1.3379 + Label L;
1.3380 + __ jccb(Assembler::notEqual, L);
1.3381 + __ movl(rdx, rcx); // if not used then remember entry in rdx
1.3382 + __ bind(L);
1.3383 + }
1.3384 + __ cmpl(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
1.3385 + __ jccb(Assembler::equal, exit); // if same object then stop searching
1.3386 + __ addl(rcx, entry_size); // otherwise advance to next entry
1.3387 + __ bind(entry);
1.3388 + __ cmpl(rcx, rbx); // check if bottom reached
1.3389 + __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1.3390 + __ bind(exit);
1.3391 + }
1.3392 +
1.3393 + __ testl(rdx, rdx); // check if a slot has been found
1.3394 + __ jccb(Assembler::notZero, allocated); // if found, continue with that one
1.3395 +
1.3396 + // allocate one if there's no free slot
1.3397 + { Label entry, loop;
1.3398 + // 1. compute new pointers // rsp: old expression stack top
1.3399 + __ movl(rdx, monitor_block_bot); // rdx: old expression stack bottom
1.3400 + __ subl(rsp, entry_size); // move expression stack top
1.3401 + __ subl(rdx, entry_size); // move expression stack bottom
1.3402 + __ movl(rcx, rsp); // set start value for copy loop
1.3403 + __ movl(monitor_block_bot, rdx); // set new monitor block top
1.3404 + __ jmp(entry);
1.3405 + // 2. move expression stack contents
1.3406 + __ bind(loop);
1.3407 + __ movl(rbx, Address(rcx, entry_size)); // load expression stack word from old location
1.3408 + __ movl(Address(rcx, 0), rbx); // and store it at new location
1.3409 + __ addl(rcx, wordSize); // advance to next word
1.3410 + __ bind(entry);
1.3411 + __ cmpl(rcx, rdx); // check if bottom reached
1.3412 + __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word
1.3413 + }
1.3414 +
1.3415 + // call run-time routine
1.3416 + // rdx: points to monitor entry
1.3417 + __ bind(allocated);
1.3418 +
1.3419 + // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
1.3420 + // The object has already been poped from the stack, so the expression stack looks correct.
1.3421 + __ increment(rsi);
1.3422 +
1.3423 + __ movl(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
1.3424 + __ lock_object(rdx);
1.3425 +
1.3426 + // check to make sure this monitor doesn't cause stack overflow after locking
1.3427 + __ save_bcp(); // in case of exception
1.3428 + __ generate_stack_overflow_check(0);
1.3429 +
1.3430 + // The bcp has already been incremented. Just need to dispatch to next instruction.
1.3431 + __ dispatch_next(vtos);
1.3432 +}
1.3433 +
1.3434 +
1.3435 +void TemplateTable::monitorexit() {
1.3436 + transition(atos, vtos);
1.3437 +
1.3438 + // check for NULL object
1.3439 + __ null_check(rax);
1.3440 +
1.3441 + const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1.3442 + const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1.3443 + const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
1.3444 + Label found;
1.3445 +
1.3446 + // find matching slot
1.3447 + { Label entry, loop;
1.3448 + __ movl(rdx, monitor_block_top); // points to current entry, starting with top-most entry
1.3449 + __ leal(rbx, monitor_block_bot); // points to word before bottom of monitor block
1.3450 + __ jmpb(entry);
1.3451 +
1.3452 + __ bind(loop);
1.3453 + __ cmpl(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
1.3454 + __ jcc(Assembler::equal, found); // if same object then stop searching
1.3455 + __ addl(rdx, entry_size); // otherwise advance to next entry
1.3456 + __ bind(entry);
1.3457 + __ cmpl(rdx, rbx); // check if bottom reached
1.3458 + __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1.3459 + }
1.3460 +
1.3461 + // error handling. Unlocking was not block-structured
1.3462 + Label end;
1.3463 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1.3464 + __ should_not_reach_here();
1.3465 +
1.3466 + // call run-time routine
1.3467 + // rcx: points to monitor entry
1.3468 + __ bind(found);
1.3469 + __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
1.3470 + __ unlock_object(rdx);
1.3471 + __ pop_ptr(rax); // discard object
1.3472 + __ bind(end);
1.3473 +}
1.3474 +
1.3475 +
1.3476 +//----------------------------------------------------------------------------------------------------
1.3477 +// Wide instructions
1.3478 +
1.3479 +void TemplateTable::wide() {
1.3480 + transition(vtos, vtos);
1.3481 + __ load_unsigned_byte(rbx, at_bcp(1));
1.3482 + __ jmp(Address(noreg, rbx, Address::times_4, int(Interpreter::_wentry_point)));
1.3483 + // Note: the rsi increment step is part of the individual wide bytecode implementations
1.3484 +}
1.3485 +
1.3486 +
1.3487 +//----------------------------------------------------------------------------------------------------
1.3488 +// Multi arrays
1.3489 +
1.3490 +void TemplateTable::multianewarray() {
1.3491 + transition(vtos, atos);
1.3492 + __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
1.3493 + // last dim is on top of stack; we want address of first one:
1.3494 + // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
1.3495 + // the latter wordSize to point to the beginning of the array.
1.3496 + __ leal( rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
1.3497 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax); // pass in rax,
1.3498 + __ load_unsigned_byte(rbx, at_bcp(3));
1.3499 + __ leal(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
1.3500 +}
1.3501 +
1.3502 +#endif /* !CC_INTERP */
