1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/src/cpu/x86/vm/templateTable_x86_64.cpp Sat Dec 01 00:00:00 2007 +0000
1.3 @@ -0,0 +1,3546 @@
1.4 +/*
1.5 + * Copyright 2003-2007 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_64.cpp.incl"
1.30 +
1.31 +#define __ _masm->
1.32 +
1.33 +// Platform-dependent initialization
1.34 +
1.35 +void TemplateTable::pd_initialize() {
1.36 + // No amd64 specific initialization
1.37 +}
1.38 +
1.39 +// Address computation: local variables
1.40 +
1.41 +static inline Address iaddress(int n) {
1.42 + return Address(r14, Interpreter::local_offset_in_bytes(n));
1.43 +}
1.44 +
1.45 +static inline Address laddress(int n) {
1.46 + return iaddress(n + 1);
1.47 +}
1.48 +
1.49 +static inline Address faddress(int n) {
1.50 + return iaddress(n);
1.51 +}
1.52 +
1.53 +static inline Address daddress(int n) {
1.54 + return laddress(n);
1.55 +}
1.56 +
1.57 +static inline Address aaddress(int n) {
1.58 + return iaddress(n);
1.59 +}
1.60 +
1.61 +static inline Address iaddress(Register r) {
1.62 + return Address(r14, r, Address::times_8, Interpreter::value_offset_in_bytes());
1.63 +}
1.64 +
1.65 +static inline Address laddress(Register r) {
1.66 + return Address(r14, r, Address::times_8, Interpreter::local_offset_in_bytes(1));
1.67 +}
1.68 +
1.69 +static inline Address faddress(Register r) {
1.70 + return iaddress(r);
1.71 +}
1.72 +
1.73 +static inline Address daddress(Register r) {
1.74 + return laddress(r);
1.75 +}
1.76 +
1.77 +static inline Address aaddress(Register r) {
1.78 + return iaddress(r);
1.79 +}
1.80 +
1.81 +static inline Address at_rsp() {
1.82 + return Address(rsp, 0);
1.83 +}
1.84 +
1.85 +// At top of Java expression stack which may be different than esp(). It
1.86 +// isn't for category 1 objects.
1.87 +static inline Address at_tos () {
1.88 + return Address(rsp, Interpreter::expr_offset_in_bytes(0));
1.89 +}
1.90 +
1.91 +static inline Address at_tos_p1() {
1.92 + return Address(rsp, Interpreter::expr_offset_in_bytes(1));
1.93 +}
1.94 +
1.95 +static inline Address at_tos_p2() {
1.96 + return Address(rsp, Interpreter::expr_offset_in_bytes(2));
1.97 +}
1.98 +
1.99 +static inline Address at_tos_p3() {
1.100 + return Address(rsp, Interpreter::expr_offset_in_bytes(3));
1.101 +}
1.102 +
1.103 +// Condition conversion
1.104 +static Assembler::Condition j_not(TemplateTable::Condition cc) {
1.105 + switch (cc) {
1.106 + case TemplateTable::equal : return Assembler::notEqual;
1.107 + case TemplateTable::not_equal : return Assembler::equal;
1.108 + case TemplateTable::less : return Assembler::greaterEqual;
1.109 + case TemplateTable::less_equal : return Assembler::greater;
1.110 + case TemplateTable::greater : return Assembler::lessEqual;
1.111 + case TemplateTable::greater_equal: return Assembler::less;
1.112 + }
1.113 + ShouldNotReachHere();
1.114 + return Assembler::zero;
1.115 +}
1.116 +
1.117 +
1.118 +// Miscelaneous helper routines
1.119 +
1.120 +Address TemplateTable::at_bcp(int offset) {
1.121 + assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
1.122 + return Address(r13, offset);
1.123 +}
1.124 +
1.125 +void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc,
1.126 + Register scratch,
1.127 + bool load_bc_into_scratch/*=true*/) {
1.128 + if (!RewriteBytecodes) {
1.129 + return;
1.130 + }
1.131 + // the pair bytecodes have already done the load.
1.132 + if (load_bc_into_scratch) {
1.133 + __ movl(bc, bytecode);
1.134 + }
1.135 + Label patch_done;
1.136 + if (JvmtiExport::can_post_breakpoint()) {
1.137 + Label fast_patch;
1.138 + // if a breakpoint is present we can't rewrite the stream directly
1.139 + __ movzbl(scratch, at_bcp(0));
1.140 + __ cmpl(scratch, Bytecodes::_breakpoint);
1.141 + __ jcc(Assembler::notEqual, fast_patch);
1.142 + __ get_method(scratch);
1.143 + // Let breakpoint table handling rewrite to quicker bytecode
1.144 + __ call_VM(noreg,
1.145 + CAST_FROM_FN_PTR(address,
1.146 + InterpreterRuntime::set_original_bytecode_at),
1.147 + scratch, r13, bc);
1.148 +#ifndef ASSERT
1.149 + __ jmpb(patch_done);
1.150 + __ bind(fast_patch);
1.151 + }
1.152 +#else
1.153 + __ jmp(patch_done);
1.154 + __ bind(fast_patch);
1.155 + }
1.156 + Label okay;
1.157 + __ load_unsigned_byte(scratch, at_bcp(0));
1.158 + __ cmpl(scratch, (int) Bytecodes::java_code(bytecode));
1.159 + __ jcc(Assembler::equal, okay);
1.160 + __ cmpl(scratch, bc);
1.161 + __ jcc(Assembler::equal, okay);
1.162 + __ stop("patching the wrong bytecode");
1.163 + __ bind(okay);
1.164 +#endif
1.165 + // patch bytecode
1.166 + __ movb(at_bcp(0), bc);
1.167 + __ bind(patch_done);
1.168 +}
1.169 +
1.170 +
1.171 +// Individual instructions
1.172 +
1.173 +void TemplateTable::nop() {
1.174 + transition(vtos, vtos);
1.175 + // nothing to do
1.176 +}
1.177 +
1.178 +void TemplateTable::shouldnotreachhere() {
1.179 + transition(vtos, vtos);
1.180 + __ stop("shouldnotreachhere bytecode");
1.181 +}
1.182 +
1.183 +void TemplateTable::aconst_null() {
1.184 + transition(vtos, atos);
1.185 + __ xorl(rax, rax);
1.186 +}
1.187 +
1.188 +void TemplateTable::iconst(int value) {
1.189 + transition(vtos, itos);
1.190 + if (value == 0) {
1.191 + __ xorl(rax, rax);
1.192 + } else {
1.193 + __ movl(rax, value);
1.194 + }
1.195 +}
1.196 +
1.197 +void TemplateTable::lconst(int value) {
1.198 + transition(vtos, ltos);
1.199 + if (value == 0) {
1.200 + __ xorl(rax, rax);
1.201 + } else {
1.202 + __ movl(rax, value);
1.203 + }
1.204 +}
1.205 +
1.206 +void TemplateTable::fconst(int value) {
1.207 + transition(vtos, ftos);
1.208 + static float one = 1.0f, two = 2.0f;
1.209 + switch (value) {
1.210 + case 0:
1.211 + __ xorps(xmm0, xmm0);
1.212 + break;
1.213 + case 1:
1.214 + __ movflt(xmm0, ExternalAddress((address) &one));
1.215 + break;
1.216 + case 2:
1.217 + __ movflt(xmm0, ExternalAddress((address) &two));
1.218 + break;
1.219 + default:
1.220 + ShouldNotReachHere();
1.221 + break;
1.222 + }
1.223 +}
1.224 +
1.225 +void TemplateTable::dconst(int value) {
1.226 + transition(vtos, dtos);
1.227 + static double one = 1.0;
1.228 + switch (value) {
1.229 + case 0:
1.230 + __ xorpd(xmm0, xmm0);
1.231 + break;
1.232 + case 1:
1.233 + __ movdbl(xmm0, ExternalAddress((address) &one));
1.234 + break;
1.235 + default:
1.236 + ShouldNotReachHere();
1.237 + break;
1.238 + }
1.239 +}
1.240 +
1.241 +void TemplateTable::bipush() {
1.242 + transition(vtos, itos);
1.243 + __ load_signed_byte(rax, at_bcp(1));
1.244 +}
1.245 +
1.246 +void TemplateTable::sipush() {
1.247 + transition(vtos, itos);
1.248 + __ load_unsigned_word(rax, at_bcp(1));
1.249 + __ bswapl(rax);
1.250 + __ sarl(rax, 16);
1.251 +}
1.252 +
1.253 +void TemplateTable::ldc(bool wide) {
1.254 + transition(vtos, vtos);
1.255 + Label call_ldc, notFloat, notClass, Done;
1.256 +
1.257 + if (wide) {
1.258 + __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
1.259 + } else {
1.260 + __ load_unsigned_byte(rbx, at_bcp(1));
1.261 + }
1.262 +
1.263 + __ get_cpool_and_tags(rcx, rax);
1.264 + const int base_offset = constantPoolOopDesc::header_size() * wordSize;
1.265 + const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
1.266 +
1.267 + // get type
1.268 + __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
1.269 +
1.270 + // unresolved string - get the resolved string
1.271 + __ cmpl(rdx, JVM_CONSTANT_UnresolvedString);
1.272 + __ jccb(Assembler::equal, call_ldc);
1.273 +
1.274 + // unresolved class - get the resolved class
1.275 + __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
1.276 + __ jccb(Assembler::equal, call_ldc);
1.277 +
1.278 + // unresolved class in error state - call into runtime to throw the error
1.279 + // from the first resolution attempt
1.280 + __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
1.281 + __ jccb(Assembler::equal, call_ldc);
1.282 +
1.283 + // resolved class - need to call vm to get java mirror of the class
1.284 + __ cmpl(rdx, JVM_CONSTANT_Class);
1.285 + __ jcc(Assembler::notEqual, notClass);
1.286 +
1.287 + __ bind(call_ldc);
1.288 + __ movl(c_rarg1, wide);
1.289 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
1.290 + __ push_ptr(rax);
1.291 + __ verify_oop(rax);
1.292 + __ jmp(Done);
1.293 +
1.294 + __ bind(notClass);
1.295 + __ cmpl(rdx, JVM_CONSTANT_Float);
1.296 + __ jccb(Assembler::notEqual, notFloat);
1.297 + // ftos
1.298 + __ movflt(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
1.299 + __ push_f();
1.300 + __ jmp(Done);
1.301 +
1.302 + __ bind(notFloat);
1.303 +#ifdef ASSERT
1.304 + {
1.305 + Label L;
1.306 + __ cmpl(rdx, JVM_CONSTANT_Integer);
1.307 + __ jcc(Assembler::equal, L);
1.308 + __ cmpl(rdx, JVM_CONSTANT_String);
1.309 + __ jcc(Assembler::equal, L);
1.310 + __ stop("unexpected tag type in ldc");
1.311 + __ bind(L);
1.312 + }
1.313 +#endif
1.314 + // atos and itos
1.315 + Label isOop;
1.316 + __ cmpl(rdx, JVM_CONSTANT_Integer);
1.317 + __ jcc(Assembler::notEqual, isOop);
1.318 + __ movl(rax, Address(rcx, rbx, Address::times_8, base_offset));
1.319 + __ push_i(rax);
1.320 + __ jmp(Done);
1.321 +
1.322 + __ bind(isOop);
1.323 + __ movq(rax, Address(rcx, rbx, Address::times_8, base_offset));
1.324 + __ push_ptr(rax);
1.325 +
1.326 + if (VerifyOops) {
1.327 + __ verify_oop(rax);
1.328 + }
1.329 +
1.330 + __ bind(Done);
1.331 +}
1.332 +
1.333 +void TemplateTable::ldc2_w() {
1.334 + transition(vtos, vtos);
1.335 + Label Long, Done;
1.336 + __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
1.337 +
1.338 + __ get_cpool_and_tags(rcx, rax);
1.339 + const int base_offset = constantPoolOopDesc::header_size() * wordSize;
1.340 + const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
1.341 +
1.342 + // get type
1.343 + __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
1.344 + JVM_CONSTANT_Double);
1.345 + __ jccb(Assembler::notEqual, Long);
1.346 + // dtos
1.347 + __ movdbl(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
1.348 + __ push_d();
1.349 + __ jmpb(Done);
1.350 +
1.351 + __ bind(Long);
1.352 + // ltos
1.353 + __ movq(rax, Address(rcx, rbx, Address::times_8, base_offset));
1.354 + __ push_l();
1.355 +
1.356 + __ bind(Done);
1.357 +}
1.358 +
1.359 +void TemplateTable::locals_index(Register reg, int offset) {
1.360 + __ load_unsigned_byte(reg, at_bcp(offset));
1.361 + __ negq(reg);
1.362 + if (TaggedStackInterpreter) __ shlq(reg, 1); // index = index*2
1.363 +}
1.364 +
1.365 +void TemplateTable::iload() {
1.366 + transition(vtos, itos);
1.367 + if (RewriteFrequentPairs) {
1.368 + Label rewrite, done;
1.369 + const Register bc = c_rarg3;
1.370 + assert(rbx != bc, "register damaged");
1.371 +
1.372 + // get next byte
1.373 + __ load_unsigned_byte(rbx,
1.374 + at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
1.375 + // if _iload, wait to rewrite to iload2. We only want to rewrite the
1.376 + // last two iloads in a pair. Comparing against fast_iload means that
1.377 + // the next bytecode is neither an iload or a caload, and therefore
1.378 + // an iload pair.
1.379 + __ cmpl(rbx, Bytecodes::_iload);
1.380 + __ jcc(Assembler::equal, done);
1.381 +
1.382 + __ cmpl(rbx, Bytecodes::_fast_iload);
1.383 + __ movl(bc, Bytecodes::_fast_iload2);
1.384 + __ jccb(Assembler::equal, rewrite);
1.385 +
1.386 + // if _caload, rewrite to fast_icaload
1.387 + __ cmpl(rbx, Bytecodes::_caload);
1.388 + __ movl(bc, Bytecodes::_fast_icaload);
1.389 + __ jccb(Assembler::equal, rewrite);
1.390 +
1.391 + // rewrite so iload doesn't check again.
1.392 + __ movl(bc, Bytecodes::_fast_iload);
1.393 +
1.394 + // rewrite
1.395 + // bc: fast bytecode
1.396 + __ bind(rewrite);
1.397 + patch_bytecode(Bytecodes::_iload, bc, rbx, false);
1.398 + __ bind(done);
1.399 + }
1.400 +
1.401 + // Get the local value into tos
1.402 + locals_index(rbx);
1.403 + __ movl(rax, iaddress(rbx));
1.404 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.405 +}
1.406 +
1.407 +void TemplateTable::fast_iload2() {
1.408 + transition(vtos, itos);
1.409 + locals_index(rbx);
1.410 + __ movl(rax, iaddress(rbx));
1.411 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.412 + __ push(itos);
1.413 + locals_index(rbx, 3);
1.414 + __ movl(rax, iaddress(rbx));
1.415 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.416 +}
1.417 +
1.418 +void TemplateTable::fast_iload() {
1.419 + transition(vtos, itos);
1.420 + locals_index(rbx);
1.421 + __ movl(rax, iaddress(rbx));
1.422 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.423 +}
1.424 +
1.425 +void TemplateTable::lload() {
1.426 + transition(vtos, ltos);
1.427 + locals_index(rbx);
1.428 + __ movq(rax, laddress(rbx));
1.429 + debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
1.430 +}
1.431 +
1.432 +void TemplateTable::fload() {
1.433 + transition(vtos, ftos);
1.434 + locals_index(rbx);
1.435 + __ movflt(xmm0, faddress(rbx));
1.436 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.437 +}
1.438 +
1.439 +void TemplateTable::dload() {
1.440 + transition(vtos, dtos);
1.441 + locals_index(rbx);
1.442 + __ movdbl(xmm0, daddress(rbx));
1.443 + debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
1.444 +}
1.445 +
1.446 +void TemplateTable::aload() {
1.447 + transition(vtos, atos);
1.448 + locals_index(rbx);
1.449 + __ movq(rax, aaddress(rbx));
1.450 + debug_only(__ verify_local_tag(frame::TagReference, rbx));
1.451 +}
1.452 +
1.453 +void TemplateTable::locals_index_wide(Register reg) {
1.454 + __ movl(reg, at_bcp(2));
1.455 + __ bswapl(reg);
1.456 + __ shrl(reg, 16);
1.457 + __ negq(reg);
1.458 + if (TaggedStackInterpreter) __ shlq(reg, 1); // index = index*2
1.459 +}
1.460 +
1.461 +void TemplateTable::wide_iload() {
1.462 + transition(vtos, itos);
1.463 + locals_index_wide(rbx);
1.464 + __ movl(rax, iaddress(rbx));
1.465 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.466 +}
1.467 +
1.468 +void TemplateTable::wide_lload() {
1.469 + transition(vtos, ltos);
1.470 + locals_index_wide(rbx);
1.471 + __ movq(rax, laddress(rbx));
1.472 + debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
1.473 +}
1.474 +
1.475 +void TemplateTable::wide_fload() {
1.476 + transition(vtos, ftos);
1.477 + locals_index_wide(rbx);
1.478 + __ movflt(xmm0, faddress(rbx));
1.479 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.480 +}
1.481 +
1.482 +void TemplateTable::wide_dload() {
1.483 + transition(vtos, dtos);
1.484 + locals_index_wide(rbx);
1.485 + __ movdbl(xmm0, daddress(rbx));
1.486 + debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
1.487 +}
1.488 +
1.489 +void TemplateTable::wide_aload() {
1.490 + transition(vtos, atos);
1.491 + locals_index_wide(rbx);
1.492 + __ movq(rax, aaddress(rbx));
1.493 + debug_only(__ verify_local_tag(frame::TagReference, rbx));
1.494 +}
1.495 +
1.496 +void TemplateTable::index_check(Register array, Register index) {
1.497 + // destroys rbx
1.498 + // check array
1.499 + __ null_check(array, arrayOopDesc::length_offset_in_bytes());
1.500 + // sign extend index for use by indexed load
1.501 + __ movslq(index, index);
1.502 + // check index
1.503 + __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
1.504 + if (index != rbx) {
1.505 + // ??? convention: move aberrant index into ebx for exception message
1.506 + assert(rbx != array, "different registers");
1.507 + __ movl(rbx, index);
1.508 + }
1.509 + __ jump_cc(Assembler::aboveEqual,
1.510 + ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
1.511 +}
1.512 +
1.513 +void TemplateTable::iaload() {
1.514 + transition(itos, itos);
1.515 + __ pop_ptr(rdx);
1.516 + // eax: index
1.517 + // rdx: array
1.518 + index_check(rdx, rax); // kills rbx
1.519 + __ movl(rax, Address(rdx, rax,
1.520 + Address::times_4,
1.521 + arrayOopDesc::base_offset_in_bytes(T_INT)));
1.522 +}
1.523 +
1.524 +void TemplateTable::laload() {
1.525 + transition(itos, ltos);
1.526 + __ pop_ptr(rdx);
1.527 + // eax: index
1.528 + // rdx: array
1.529 + index_check(rdx, rax); // kills rbx
1.530 + __ movq(rax, Address(rdx, rbx,
1.531 + Address::times_8,
1.532 + arrayOopDesc::base_offset_in_bytes(T_LONG)));
1.533 +}
1.534 +
1.535 +void TemplateTable::faload() {
1.536 + transition(itos, ftos);
1.537 + __ pop_ptr(rdx);
1.538 + // eax: index
1.539 + // rdx: array
1.540 + index_check(rdx, rax); // kills rbx
1.541 + __ movflt(xmm0, Address(rdx, rax,
1.542 + Address::times_4,
1.543 + arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1.544 +}
1.545 +
1.546 +void TemplateTable::daload() {
1.547 + transition(itos, dtos);
1.548 + __ pop_ptr(rdx);
1.549 + // eax: index
1.550 + // rdx: array
1.551 + index_check(rdx, rax); // kills rbx
1.552 + __ movdbl(xmm0, Address(rdx, rax,
1.553 + Address::times_8,
1.554 + arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1.555 +}
1.556 +
1.557 +void TemplateTable::aaload() {
1.558 + transition(itos, atos);
1.559 + __ pop_ptr(rdx);
1.560 + // eax: index
1.561 + // rdx: array
1.562 + index_check(rdx, rax); // kills rbx
1.563 + __ movq(rax, Address(rdx, rax,
1.564 + Address::times_8,
1.565 + arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1.566 +}
1.567 +
1.568 +void TemplateTable::baload() {
1.569 + transition(itos, itos);
1.570 + __ pop_ptr(rdx);
1.571 + // eax: index
1.572 + // rdx: array
1.573 + index_check(rdx, rax); // kills rbx
1.574 + __ load_signed_byte(rax,
1.575 + Address(rdx, rax,
1.576 + Address::times_1,
1.577 + arrayOopDesc::base_offset_in_bytes(T_BYTE)));
1.578 +}
1.579 +
1.580 +void TemplateTable::caload() {
1.581 + transition(itos, itos);
1.582 + __ pop_ptr(rdx);
1.583 + // eax: index
1.584 + // rdx: array
1.585 + index_check(rdx, rax); // kills rbx
1.586 + __ load_unsigned_word(rax,
1.587 + Address(rdx, rax,
1.588 + Address::times_2,
1.589 + arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1.590 +}
1.591 +
1.592 +// iload followed by caload frequent pair
1.593 +void TemplateTable::fast_icaload() {
1.594 + transition(vtos, itos);
1.595 + // load index out of locals
1.596 + locals_index(rbx);
1.597 + __ movl(rax, iaddress(rbx));
1.598 + debug_only(__ verify_local_tag(frame::TagValue, rbx));
1.599 +
1.600 + // eax: index
1.601 + // rdx: array
1.602 + __ pop_ptr(rdx);
1.603 + index_check(rdx, rax); // kills rbx
1.604 + __ load_unsigned_word(rax,
1.605 + Address(rdx, rax,
1.606 + Address::times_2,
1.607 + arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1.608 +}
1.609 +
1.610 +void TemplateTable::saload() {
1.611 + transition(itos, itos);
1.612 + __ pop_ptr(rdx);
1.613 + // eax: index
1.614 + // rdx: array
1.615 + index_check(rdx, rax); // kills rbx
1.616 + __ load_signed_word(rax,
1.617 + Address(rdx, rax,
1.618 + Address::times_2,
1.619 + arrayOopDesc::base_offset_in_bytes(T_SHORT)));
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 +void TemplateTable::lload(int n) {
1.629 + transition(vtos, ltos);
1.630 + __ movq(rax, laddress(n));
1.631 + debug_only(__ verify_local_tag(frame::TagCategory2, n));
1.632 +}
1.633 +
1.634 +void TemplateTable::fload(int n) {
1.635 + transition(vtos, ftos);
1.636 + __ movflt(xmm0, faddress(n));
1.637 + debug_only(__ verify_local_tag(frame::TagValue, n));
1.638 +}
1.639 +
1.640 +void TemplateTable::dload(int n) {
1.641 + transition(vtos, dtos);
1.642 + __ movdbl(xmm0, daddress(n));
1.643 + debug_only(__ verify_local_tag(frame::TagCategory2, n));
1.644 +}
1.645 +
1.646 +void TemplateTable::aload(int n) {
1.647 + transition(vtos, atos);
1.648 + __ movq(rax, aaddress(n));
1.649 + debug_only(__ verify_local_tag(frame::TagReference, n));
1.650 +}
1.651 +
1.652 +void TemplateTable::aload_0() {
1.653 + transition(vtos, atos);
1.654 + // According to bytecode histograms, the pairs:
1.655 + //
1.656 + // _aload_0, _fast_igetfield
1.657 + // _aload_0, _fast_agetfield
1.658 + // _aload_0, _fast_fgetfield
1.659 + //
1.660 + // occur frequently. If RewriteFrequentPairs is set, the (slow)
1.661 + // _aload_0 bytecode checks if the next bytecode is either
1.662 + // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
1.663 + // rewrites the current bytecode into a pair bytecode; otherwise it
1.664 + // rewrites the current bytecode into _fast_aload_0 that doesn't do
1.665 + // the pair check anymore.
1.666 + //
1.667 + // Note: If the next bytecode is _getfield, the rewrite must be
1.668 + // delayed, otherwise we may miss an opportunity for a pair.
1.669 + //
1.670 + // Also rewrite frequent pairs
1.671 + // aload_0, aload_1
1.672 + // aload_0, iload_1
1.673 + // These bytecodes with a small amount of code are most profitable
1.674 + // to rewrite
1.675 + if (RewriteFrequentPairs) {
1.676 + Label rewrite, done;
1.677 + const Register bc = c_rarg3;
1.678 + assert(rbx != bc, "register damaged");
1.679 + // get next byte
1.680 + __ load_unsigned_byte(rbx,
1.681 + at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
1.682 +
1.683 + // do actual aload_0
1.684 + aload(0);
1.685 +
1.686 + // if _getfield then wait with rewrite
1.687 + __ cmpl(rbx, Bytecodes::_getfield);
1.688 + __ jcc(Assembler::equal, done);
1.689 +
1.690 + // if _igetfield then reqrite to _fast_iaccess_0
1.691 + assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) ==
1.692 + Bytecodes::_aload_0,
1.693 + "fix bytecode definition");
1.694 + __ cmpl(rbx, Bytecodes::_fast_igetfield);
1.695 + __ movl(bc, Bytecodes::_fast_iaccess_0);
1.696 + __ jccb(Assembler::equal, rewrite);
1.697 +
1.698 + // if _agetfield then reqrite to _fast_aaccess_0
1.699 + assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) ==
1.700 + Bytecodes::_aload_0,
1.701 + "fix bytecode definition");
1.702 + __ cmpl(rbx, Bytecodes::_fast_agetfield);
1.703 + __ movl(bc, Bytecodes::_fast_aaccess_0);
1.704 + __ jccb(Assembler::equal, rewrite);
1.705 +
1.706 + // if _fgetfield then reqrite to _fast_faccess_0
1.707 + assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) ==
1.708 + Bytecodes::_aload_0,
1.709 + "fix bytecode definition");
1.710 + __ cmpl(rbx, Bytecodes::_fast_fgetfield);
1.711 + __ movl(bc, Bytecodes::_fast_faccess_0);
1.712 + __ jccb(Assembler::equal, rewrite);
1.713 +
1.714 + // else rewrite to _fast_aload0
1.715 + assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) ==
1.716 + Bytecodes::_aload_0,
1.717 + "fix bytecode definition");
1.718 + __ movl(bc, Bytecodes::_fast_aload_0);
1.719 +
1.720 + // rewrite
1.721 + // bc: fast bytecode
1.722 + __ bind(rewrite);
1.723 + patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
1.724 +
1.725 + __ bind(done);
1.726 + } else {
1.727 + aload(0);
1.728 + }
1.729 +}
1.730 +
1.731 +void TemplateTable::istore() {
1.732 + transition(itos, vtos);
1.733 + locals_index(rbx);
1.734 + __ movl(iaddress(rbx), rax);
1.735 + __ tag_local(frame::TagValue, rbx);
1.736 +}
1.737 +
1.738 +void TemplateTable::lstore() {
1.739 + transition(ltos, vtos);
1.740 + locals_index(rbx);
1.741 + __ movq(laddress(rbx), rax);
1.742 + __ tag_local(frame::TagCategory2, rbx);
1.743 +}
1.744 +
1.745 +void TemplateTable::fstore() {
1.746 + transition(ftos, vtos);
1.747 + locals_index(rbx);
1.748 + __ movflt(faddress(rbx), xmm0);
1.749 + __ tag_local(frame::TagValue, rbx);
1.750 +}
1.751 +
1.752 +void TemplateTable::dstore() {
1.753 + transition(dtos, vtos);
1.754 + locals_index(rbx);
1.755 + __ movdbl(daddress(rbx), xmm0);
1.756 + __ tag_local(frame::TagCategory2, rbx);
1.757 +}
1.758 +
1.759 +void TemplateTable::astore() {
1.760 + transition(vtos, vtos);
1.761 + __ pop_ptr(rax, rdx); // will need to pop tag too
1.762 + locals_index(rbx);
1.763 + __ movq(aaddress(rbx), rax);
1.764 + __ tag_local(rdx, rbx); // store tag from stack, might be returnAddr
1.765 +}
1.766 +
1.767 +void TemplateTable::wide_istore() {
1.768 + transition(vtos, vtos);
1.769 + __ pop_i();
1.770 + locals_index_wide(rbx);
1.771 + __ movl(iaddress(rbx), rax);
1.772 + __ tag_local(frame::TagValue, rbx);
1.773 +}
1.774 +
1.775 +void TemplateTable::wide_lstore() {
1.776 + transition(vtos, vtos);
1.777 + __ pop_l();
1.778 + locals_index_wide(rbx);
1.779 + __ movq(laddress(rbx), rax);
1.780 + __ tag_local(frame::TagCategory2, rbx);
1.781 +}
1.782 +
1.783 +void TemplateTable::wide_fstore() {
1.784 + transition(vtos, vtos);
1.785 + __ pop_f();
1.786 + locals_index_wide(rbx);
1.787 + __ movflt(faddress(rbx), xmm0);
1.788 + __ tag_local(frame::TagValue, rbx);
1.789 +}
1.790 +
1.791 +void TemplateTable::wide_dstore() {
1.792 + transition(vtos, vtos);
1.793 + __ pop_d();
1.794 + locals_index_wide(rbx);
1.795 + __ movdbl(daddress(rbx), xmm0);
1.796 + __ tag_local(frame::TagCategory2, rbx);
1.797 +}
1.798 +
1.799 +void TemplateTable::wide_astore() {
1.800 + transition(vtos, vtos);
1.801 + __ pop_ptr(rax, rdx); // will need to pop tag too
1.802 + locals_index_wide(rbx);
1.803 + __ movq(aaddress(rbx), rax);
1.804 + __ tag_local(rdx, rbx); // store tag from stack, might be returnAddr
1.805 +}
1.806 +
1.807 +void TemplateTable::iastore() {
1.808 + transition(itos, vtos);
1.809 + __ pop_i(rbx);
1.810 + __ pop_ptr(rdx);
1.811 + // eax: value
1.812 + // ebx: index
1.813 + // rdx: array
1.814 + index_check(rdx, rbx); // prefer index in ebx
1.815 + __ movl(Address(rdx, rbx,
1.816 + Address::times_4,
1.817 + arrayOopDesc::base_offset_in_bytes(T_INT)),
1.818 + rax);
1.819 +}
1.820 +
1.821 +void TemplateTable::lastore() {
1.822 + transition(ltos, vtos);
1.823 + __ pop_i(rbx);
1.824 + __ pop_ptr(rdx);
1.825 + // rax: value
1.826 + // ebx: index
1.827 + // rdx: array
1.828 + index_check(rdx, rbx); // prefer index in ebx
1.829 + __ movq(Address(rdx, rbx,
1.830 + Address::times_8,
1.831 + arrayOopDesc::base_offset_in_bytes(T_LONG)),
1.832 + rax);
1.833 +}
1.834 +
1.835 +void TemplateTable::fastore() {
1.836 + transition(ftos, vtos);
1.837 + __ pop_i(rbx);
1.838 + __ pop_ptr(rdx);
1.839 + // xmm0: value
1.840 + // ebx: index
1.841 + // rdx: array
1.842 + index_check(rdx, rbx); // prefer index in ebx
1.843 + __ movflt(Address(rdx, rbx,
1.844 + Address::times_4,
1.845 + arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
1.846 + xmm0);
1.847 +}
1.848 +
1.849 +void TemplateTable::dastore() {
1.850 + transition(dtos, vtos);
1.851 + __ pop_i(rbx);
1.852 + __ pop_ptr(rdx);
1.853 + // xmm0: value
1.854 + // ebx: index
1.855 + // rdx: array
1.856 + index_check(rdx, rbx); // prefer index in ebx
1.857 + __ movdbl(Address(rdx, rbx,
1.858 + Address::times_8,
1.859 + arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
1.860 + xmm0);
1.861 +}
1.862 +
1.863 +void TemplateTable::aastore() {
1.864 + Label is_null, ok_is_subtype, done;
1.865 + transition(vtos, vtos);
1.866 + // stack: ..., array, index, value
1.867 + __ movq(rax, at_tos()); // value
1.868 + __ movl(rcx, at_tos_p1()); // index
1.869 + __ movq(rdx, at_tos_p2()); // array
1.870 + index_check(rdx, rcx); // kills rbx
1.871 + // do array store check - check for NULL value first
1.872 + __ testq(rax, rax);
1.873 + __ jcc(Assembler::zero, is_null);
1.874 +
1.875 + // Move subklass into rbx
1.876 + __ movq(rbx, Address(rax, oopDesc::klass_offset_in_bytes()));
1.877 + // Move superklass into rax
1.878 + __ movq(rax, Address(rdx, oopDesc::klass_offset_in_bytes()));
1.879 + __ movq(rax, Address(rax,
1.880 + sizeof(oopDesc) +
1.881 + objArrayKlass::element_klass_offset_in_bytes()));
1.882 + // Compress array + index*8 + 12 into a single register. Frees rcx.
1.883 + __ leaq(rdx, Address(rdx, rcx,
1.884 + Address::times_8,
1.885 + arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1.886 +
1.887 + // Generate subtype check. Blows rcx, rdi
1.888 + // Superklass in rax. Subklass in rbx.
1.889 + __ gen_subtype_check(rbx, ok_is_subtype);
1.890 +
1.891 + // Come here on failure
1.892 + // object is at TOS
1.893 + __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1.894 +
1.895 + // Come here on success
1.896 + __ bind(ok_is_subtype);
1.897 + __ movq(rax, at_tos()); // Value
1.898 + __ movq(Address(rdx, 0), rax);
1.899 + __ store_check(rdx);
1.900 + __ jmp(done);
1.901 +
1.902 + // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1.903 + __ bind(is_null);
1.904 + __ profile_null_seen(rbx);
1.905 + __ movq(Address(rdx, rcx,
1.906 + Address::times_8,
1.907 + arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
1.908 + rax);
1.909 +
1.910 + // Pop stack arguments
1.911 + __ bind(done);
1.912 + __ addq(rsp, 3 * Interpreter::stackElementSize());
1.913 +}
1.914 +
1.915 +void TemplateTable::bastore() {
1.916 + transition(itos, vtos);
1.917 + __ pop_i(rbx);
1.918 + __ pop_ptr(rdx);
1.919 + // eax: value
1.920 + // ebx: index
1.921 + // rdx: array
1.922 + index_check(rdx, rbx); // prefer index in ebx
1.923 + __ movb(Address(rdx, rbx,
1.924 + Address::times_1,
1.925 + arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1.926 + rax);
1.927 +}
1.928 +
1.929 +void TemplateTable::castore() {
1.930 + transition(itos, vtos);
1.931 + __ pop_i(rbx);
1.932 + __ pop_ptr(rdx);
1.933 + // eax: value
1.934 + // ebx: index
1.935 + // rdx: array
1.936 + index_check(rdx, rbx); // prefer index in ebx
1.937 + __ movw(Address(rdx, rbx,
1.938 + Address::times_2,
1.939 + arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1.940 + rax);
1.941 +}
1.942 +
1.943 +void TemplateTable::sastore() {
1.944 + castore();
1.945 +}
1.946 +
1.947 +void TemplateTable::istore(int n) {
1.948 + transition(itos, vtos);
1.949 + __ movl(iaddress(n), rax);
1.950 + __ tag_local(frame::TagValue, n);
1.951 +}
1.952 +
1.953 +void TemplateTable::lstore(int n) {
1.954 + transition(ltos, vtos);
1.955 + __ movq(laddress(n), rax);
1.956 + __ tag_local(frame::TagCategory2, n);
1.957 +}
1.958 +
1.959 +void TemplateTable::fstore(int n) {
1.960 + transition(ftos, vtos);
1.961 + __ movflt(faddress(n), xmm0);
1.962 + __ tag_local(frame::TagValue, n);
1.963 +}
1.964 +
1.965 +void TemplateTable::dstore(int n) {
1.966 + transition(dtos, vtos);
1.967 + __ movdbl(daddress(n), xmm0);
1.968 + __ tag_local(frame::TagCategory2, n);
1.969 +}
1.970 +
1.971 +void TemplateTable::astore(int n) {
1.972 + transition(vtos, vtos);
1.973 + __ pop_ptr(rax, rdx);
1.974 + __ movq(aaddress(n), rax);
1.975 + __ tag_local(rdx, n);
1.976 +}
1.977 +
1.978 +void TemplateTable::pop() {
1.979 + transition(vtos, vtos);
1.980 + __ addq(rsp, Interpreter::stackElementSize());
1.981 +}
1.982 +
1.983 +void TemplateTable::pop2() {
1.984 + transition(vtos, vtos);
1.985 + __ addq(rsp, 2 * Interpreter::stackElementSize());
1.986 +}
1.987 +
1.988 +void TemplateTable::dup() {
1.989 + transition(vtos, vtos);
1.990 + __ load_ptr_and_tag(0, rax, rdx);
1.991 + __ push_ptr(rax, rdx);
1.992 + // stack: ..., a, a
1.993 +}
1.994 +
1.995 +void TemplateTable::dup_x1() {
1.996 + transition(vtos, vtos);
1.997 + // stack: ..., a, b
1.998 + __ load_ptr_and_tag(0, rax, rdx); // load b
1.999 + __ load_ptr_and_tag(1, rcx, rbx); // load a
1.1000 + __ store_ptr_and_tag(1, rax, rdx); // store b
1.1001 + __ store_ptr_and_tag(0, rcx, rbx); // store a
1.1002 + __ push_ptr(rax, rdx); // push b
1.1003 + // stack: ..., b, a, b
1.1004 +}
1.1005 +
1.1006 +void TemplateTable::dup_x2() {
1.1007 + transition(vtos, vtos);
1.1008 + // stack: ..., a, b, c
1.1009 + __ load_ptr_and_tag(0, rax, rdx); // load c
1.1010 + __ load_ptr_and_tag(2, rcx, rbx); // load a
1.1011 + __ store_ptr_and_tag(2, rax, rdx); // store c in a
1.1012 + __ push_ptr(rax, rdx); // push c
1.1013 + // stack: ..., c, b, c, c
1.1014 + __ load_ptr_and_tag(2, rax, rdx); // load b
1.1015 + __ store_ptr_and_tag(2, rcx, rbx); // store a in b
1.1016 + // stack: ..., c, a, c, c
1.1017 + __ store_ptr_and_tag(1, rax, rdx); // store b in c
1.1018 + // stack: ..., c, a, b, c
1.1019 +}
1.1020 +
1.1021 +void TemplateTable::dup2() {
1.1022 + transition(vtos, vtos);
1.1023 + // stack: ..., a, b
1.1024 + __ load_ptr_and_tag(1, rax, rdx); // load a
1.1025 + __ push_ptr(rax, rdx); // push a
1.1026 + __ load_ptr_and_tag(1, rax, rdx); // load b
1.1027 + __ push_ptr(rax, rdx); // push b
1.1028 + // stack: ..., a, b, a, b
1.1029 +}
1.1030 +
1.1031 +void TemplateTable::dup2_x1() {
1.1032 + transition(vtos, vtos);
1.1033 + // stack: ..., a, b, c
1.1034 + __ load_ptr_and_tag(0, rcx, rbx); // load c
1.1035 + __ load_ptr_and_tag(1, rax, rdx); // load b
1.1036 + __ push_ptr(rax, rdx); // push b
1.1037 + __ push_ptr(rcx, rbx); // push c
1.1038 + // stack: ..., a, b, c, b, c
1.1039 + __ store_ptr_and_tag(3, rcx, rbx); // store c in b
1.1040 + // stack: ..., a, c, c, b, c
1.1041 + __ load_ptr_and_tag(4, rcx, rbx); // load a
1.1042 + __ store_ptr_and_tag(2, rcx, rbx); // store a in 2nd c
1.1043 + // stack: ..., a, c, a, b, c
1.1044 + __ store_ptr_and_tag(4, rax, rdx); // store b in a
1.1045 + // stack: ..., b, c, a, b, c
1.1046 +}
1.1047 +
1.1048 +void TemplateTable::dup2_x2() {
1.1049 + transition(vtos, vtos);
1.1050 + // stack: ..., a, b, c, d
1.1051 + __ load_ptr_and_tag(0, rcx, rbx); // load d
1.1052 + __ load_ptr_and_tag(1, rax, rdx); // load c
1.1053 + __ push_ptr(rax, rdx); // push c
1.1054 + __ push_ptr(rcx, rbx); // push d
1.1055 + // stack: ..., a, b, c, d, c, d
1.1056 + __ load_ptr_and_tag(4, rax, rdx); // load b
1.1057 + __ store_ptr_and_tag(2, rax, rdx); // store b in d
1.1058 + __ store_ptr_and_tag(4, rcx, rbx); // store d in b
1.1059 + // stack: ..., a, d, c, b, c, d
1.1060 + __ load_ptr_and_tag(5, rcx, rbx); // load a
1.1061 + __ load_ptr_and_tag(3, rax, rdx); // load c
1.1062 + __ store_ptr_and_tag(3, rcx, rbx); // store a in c
1.1063 + __ store_ptr_and_tag(5, rax, rdx); // store c in a
1.1064 + // stack: ..., c, d, a, b, c, d
1.1065 +}
1.1066 +
1.1067 +void TemplateTable::swap() {
1.1068 + transition(vtos, vtos);
1.1069 + // stack: ..., a, b
1.1070 + __ load_ptr_and_tag(1, rcx, rbx); // load a
1.1071 + __ load_ptr_and_tag(0, rax, rdx); // load b
1.1072 + __ store_ptr_and_tag(0, rcx, rbx); // store a in b
1.1073 + __ store_ptr_and_tag(1, rax, rdx); // store b in a
1.1074 + // stack: ..., b, a
1.1075 +}
1.1076 +
1.1077 +void TemplateTable::iop2(Operation op) {
1.1078 + transition(itos, itos);
1.1079 + switch (op) {
1.1080 + case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1.1081 + case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1.1082 + case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1.1083 + case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1.1084 + case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1.1085 + case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1.1086 + case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1.1087 + case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1.1088 + case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1.1089 + default : ShouldNotReachHere();
1.1090 + }
1.1091 +}
1.1092 +
1.1093 +void TemplateTable::lop2(Operation op) {
1.1094 + transition(ltos, ltos);
1.1095 + switch (op) {
1.1096 + case add : __ pop_l(rdx); __ addq (rax, rdx); break;
1.1097 + case sub : __ movq(rdx, rax); __ pop_l(rax); __ subq (rax, rdx); break;
1.1098 + case _and : __ pop_l(rdx); __ andq (rax, rdx); break;
1.1099 + case _or : __ pop_l(rdx); __ orq (rax, rdx); break;
1.1100 + case _xor : __ pop_l(rdx); __ xorq (rax, rdx); break;
1.1101 + default : ShouldNotReachHere();
1.1102 + }
1.1103 +}
1.1104 +
1.1105 +void TemplateTable::idiv() {
1.1106 + transition(itos, itos);
1.1107 + __ movl(rcx, rax);
1.1108 + __ pop_i(rax);
1.1109 + // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1.1110 + // they are not equal, one could do a normal division (no correction
1.1111 + // needed), which may speed up this implementation for the common case.
1.1112 + // (see also JVM spec., p.243 & p.271)
1.1113 + __ corrected_idivl(rcx);
1.1114 +}
1.1115 +
1.1116 +void TemplateTable::irem() {
1.1117 + transition(itos, itos);
1.1118 + __ movl(rcx, rax);
1.1119 + __ pop_i(rax);
1.1120 + // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1.1121 + // they are not equal, one could do a normal division (no correction
1.1122 + // needed), which may speed up this implementation for the common case.
1.1123 + // (see also JVM spec., p.243 & p.271)
1.1124 + __ corrected_idivl(rcx);
1.1125 + __ movl(rax, rdx);
1.1126 +}
1.1127 +
1.1128 +void TemplateTable::lmul() {
1.1129 + transition(ltos, ltos);
1.1130 + __ pop_l(rdx);
1.1131 + __ imulq(rax, rdx);
1.1132 +}
1.1133 +
1.1134 +void TemplateTable::ldiv() {
1.1135 + transition(ltos, ltos);
1.1136 + __ movq(rcx, rax);
1.1137 + __ pop_l(rax);
1.1138 + // generate explicit div0 check
1.1139 + __ testq(rcx, rcx);
1.1140 + __ jump_cc(Assembler::zero,
1.1141 + ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1.1142 + // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1.1143 + // they are not equal, one could do a normal division (no correction
1.1144 + // needed), which may speed up this implementation for the common case.
1.1145 + // (see also JVM spec., p.243 & p.271)
1.1146 + __ corrected_idivq(rcx); // kills rbx
1.1147 +}
1.1148 +
1.1149 +void TemplateTable::lrem() {
1.1150 + transition(ltos, ltos);
1.1151 + __ movq(rcx, rax);
1.1152 + __ pop_l(rax);
1.1153 + __ testq(rcx, rcx);
1.1154 + __ jump_cc(Assembler::zero,
1.1155 + ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1.1156 + // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1.1157 + // they are not equal, one could do a normal division (no correction
1.1158 + // needed), which may speed up this implementation for the common case.
1.1159 + // (see also JVM spec., p.243 & p.271)
1.1160 + __ corrected_idivq(rcx); // kills rbx
1.1161 + __ movq(rax, rdx);
1.1162 +}
1.1163 +
1.1164 +void TemplateTable::lshl() {
1.1165 + transition(itos, ltos);
1.1166 + __ movl(rcx, rax); // get shift count
1.1167 + __ pop_l(rax); // get shift value
1.1168 + __ shlq(rax);
1.1169 +}
1.1170 +
1.1171 +void TemplateTable::lshr() {
1.1172 + transition(itos, ltos);
1.1173 + __ movl(rcx, rax); // get shift count
1.1174 + __ pop_l(rax); // get shift value
1.1175 + __ sarq(rax);
1.1176 +}
1.1177 +
1.1178 +void TemplateTable::lushr() {
1.1179 + transition(itos, ltos);
1.1180 + __ movl(rcx, rax); // get shift count
1.1181 + __ pop_l(rax); // get shift value
1.1182 + __ shrq(rax);
1.1183 +}
1.1184 +
1.1185 +void TemplateTable::fop2(Operation op) {
1.1186 + transition(ftos, ftos);
1.1187 + switch (op) {
1.1188 + case add:
1.1189 + __ addss(xmm0, at_rsp());
1.1190 + __ addq(rsp, Interpreter::stackElementSize());
1.1191 + break;
1.1192 + case sub:
1.1193 + __ movflt(xmm1, xmm0);
1.1194 + __ pop_f(xmm0);
1.1195 + __ subss(xmm0, xmm1);
1.1196 + break;
1.1197 + case mul:
1.1198 + __ mulss(xmm0, at_rsp());
1.1199 + __ addq(rsp, Interpreter::stackElementSize());
1.1200 + break;
1.1201 + case div:
1.1202 + __ movflt(xmm1, xmm0);
1.1203 + __ pop_f(xmm0);
1.1204 + __ divss(xmm0, xmm1);
1.1205 + break;
1.1206 + case rem:
1.1207 + __ movflt(xmm1, xmm0);
1.1208 + __ pop_f(xmm0);
1.1209 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1.1210 + break;
1.1211 + default:
1.1212 + ShouldNotReachHere();
1.1213 + break;
1.1214 + }
1.1215 +}
1.1216 +
1.1217 +void TemplateTable::dop2(Operation op) {
1.1218 + transition(dtos, dtos);
1.1219 + switch (op) {
1.1220 + case add:
1.1221 + __ addsd(xmm0, at_rsp());
1.1222 + __ addq(rsp, 2 * Interpreter::stackElementSize());
1.1223 + break;
1.1224 + case sub:
1.1225 + __ movdbl(xmm1, xmm0);
1.1226 + __ pop_d(xmm0);
1.1227 + __ subsd(xmm0, xmm1);
1.1228 + break;
1.1229 + case mul:
1.1230 + __ mulsd(xmm0, at_rsp());
1.1231 + __ addq(rsp, 2 * Interpreter::stackElementSize());
1.1232 + break;
1.1233 + case div:
1.1234 + __ movdbl(xmm1, xmm0);
1.1235 + __ pop_d(xmm0);
1.1236 + __ divsd(xmm0, xmm1);
1.1237 + break;
1.1238 + case rem:
1.1239 + __ movdbl(xmm1, xmm0);
1.1240 + __ pop_d(xmm0);
1.1241 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1.1242 + break;
1.1243 + default:
1.1244 + ShouldNotReachHere();
1.1245 + break;
1.1246 + }
1.1247 +}
1.1248 +
1.1249 +void TemplateTable::ineg() {
1.1250 + transition(itos, itos);
1.1251 + __ negl(rax);
1.1252 +}
1.1253 +
1.1254 +void TemplateTable::lneg() {
1.1255 + transition(ltos, ltos);
1.1256 + __ negq(rax);
1.1257 +}
1.1258 +
1.1259 +// Note: 'double' and 'long long' have 32-bits alignment on x86.
1.1260 +static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1.1261 + // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1.1262 + // of 128-bits operands for SSE instructions.
1.1263 + jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1.1264 + // Store the value to a 128-bits operand.
1.1265 + operand[0] = lo;
1.1266 + operand[1] = hi;
1.1267 + return operand;
1.1268 +}
1.1269 +
1.1270 +// Buffer for 128-bits masks used by SSE instructions.
1.1271 +static jlong float_signflip_pool[2*2];
1.1272 +static jlong double_signflip_pool[2*2];
1.1273 +
1.1274 +void TemplateTable::fneg() {
1.1275 + transition(ftos, ftos);
1.1276 + static jlong *float_signflip = double_quadword(&float_signflip_pool[1], 0x8000000080000000, 0x8000000080000000);
1.1277 + __ xorps(xmm0, ExternalAddress((address) float_signflip));
1.1278 +}
1.1279 +
1.1280 +void TemplateTable::dneg() {
1.1281 + transition(dtos, dtos);
1.1282 + static jlong *double_signflip = double_quadword(&double_signflip_pool[1], 0x8000000000000000, 0x8000000000000000);
1.1283 + __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1.1284 +}
1.1285 +
1.1286 +void TemplateTable::iinc() {
1.1287 + transition(vtos, vtos);
1.1288 + __ load_signed_byte(rdx, at_bcp(2)); // get constant
1.1289 + locals_index(rbx);
1.1290 + __ addl(iaddress(rbx), rdx);
1.1291 +}
1.1292 +
1.1293 +void TemplateTable::wide_iinc() {
1.1294 + transition(vtos, vtos);
1.1295 + __ movl(rdx, at_bcp(4)); // get constant
1.1296 + locals_index_wide(rbx);
1.1297 + __ bswapl(rdx); // swap bytes & sign-extend constant
1.1298 + __ sarl(rdx, 16);
1.1299 + __ addl(iaddress(rbx), rdx);
1.1300 + // Note: should probably use only one movl to get both
1.1301 + // the index and the constant -> fix this
1.1302 +}
1.1303 +
1.1304 +void TemplateTable::convert() {
1.1305 + // Checking
1.1306 +#ifdef ASSERT
1.1307 + {
1.1308 + TosState tos_in = ilgl;
1.1309 + TosState tos_out = ilgl;
1.1310 + switch (bytecode()) {
1.1311 + case Bytecodes::_i2l: // fall through
1.1312 + case Bytecodes::_i2f: // fall through
1.1313 + case Bytecodes::_i2d: // fall through
1.1314 + case Bytecodes::_i2b: // fall through
1.1315 + case Bytecodes::_i2c: // fall through
1.1316 + case Bytecodes::_i2s: tos_in = itos; break;
1.1317 + case Bytecodes::_l2i: // fall through
1.1318 + case Bytecodes::_l2f: // fall through
1.1319 + case Bytecodes::_l2d: tos_in = ltos; break;
1.1320 + case Bytecodes::_f2i: // fall through
1.1321 + case Bytecodes::_f2l: // fall through
1.1322 + case Bytecodes::_f2d: tos_in = ftos; break;
1.1323 + case Bytecodes::_d2i: // fall through
1.1324 + case Bytecodes::_d2l: // fall through
1.1325 + case Bytecodes::_d2f: tos_in = dtos; break;
1.1326 + default : ShouldNotReachHere();
1.1327 + }
1.1328 + switch (bytecode()) {
1.1329 + case Bytecodes::_l2i: // fall through
1.1330 + case Bytecodes::_f2i: // fall through
1.1331 + case Bytecodes::_d2i: // fall through
1.1332 + case Bytecodes::_i2b: // fall through
1.1333 + case Bytecodes::_i2c: // fall through
1.1334 + case Bytecodes::_i2s: tos_out = itos; break;
1.1335 + case Bytecodes::_i2l: // fall through
1.1336 + case Bytecodes::_f2l: // fall through
1.1337 + case Bytecodes::_d2l: tos_out = ltos; break;
1.1338 + case Bytecodes::_i2f: // fall through
1.1339 + case Bytecodes::_l2f: // fall through
1.1340 + case Bytecodes::_d2f: tos_out = ftos; break;
1.1341 + case Bytecodes::_i2d: // fall through
1.1342 + case Bytecodes::_l2d: // fall through
1.1343 + case Bytecodes::_f2d: tos_out = dtos; break;
1.1344 + default : ShouldNotReachHere();
1.1345 + }
1.1346 + transition(tos_in, tos_out);
1.1347 + }
1.1348 +#endif // ASSERT
1.1349 +
1.1350 + static const int64_t is_nan = 0x8000000000000000L;
1.1351 +
1.1352 + // Conversion
1.1353 + switch (bytecode()) {
1.1354 + case Bytecodes::_i2l:
1.1355 + __ movslq(rax, rax);
1.1356 + break;
1.1357 + case Bytecodes::_i2f:
1.1358 + __ cvtsi2ssl(xmm0, rax);
1.1359 + break;
1.1360 + case Bytecodes::_i2d:
1.1361 + __ cvtsi2sdl(xmm0, rax);
1.1362 + break;
1.1363 + case Bytecodes::_i2b:
1.1364 + __ movsbl(rax, rax);
1.1365 + break;
1.1366 + case Bytecodes::_i2c:
1.1367 + __ movzwl(rax, rax);
1.1368 + break;
1.1369 + case Bytecodes::_i2s:
1.1370 + __ movswl(rax, rax);
1.1371 + break;
1.1372 + case Bytecodes::_l2i:
1.1373 + __ movl(rax, rax);
1.1374 + break;
1.1375 + case Bytecodes::_l2f:
1.1376 + __ cvtsi2ssq(xmm0, rax);
1.1377 + break;
1.1378 + case Bytecodes::_l2d:
1.1379 + __ cvtsi2sdq(xmm0, rax);
1.1380 + break;
1.1381 + case Bytecodes::_f2i:
1.1382 + {
1.1383 + Label L;
1.1384 + __ cvttss2sil(rax, xmm0);
1.1385 + __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1.1386 + __ jcc(Assembler::notEqual, L);
1.1387 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1.1388 + __ bind(L);
1.1389 + }
1.1390 + break;
1.1391 + case Bytecodes::_f2l:
1.1392 + {
1.1393 + Label L;
1.1394 + __ cvttss2siq(rax, xmm0);
1.1395 + // NaN or overflow/underflow?
1.1396 + __ cmp64(rax, ExternalAddress((address) &is_nan));
1.1397 + __ jcc(Assembler::notEqual, L);
1.1398 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1.1399 + __ bind(L);
1.1400 + }
1.1401 + break;
1.1402 + case Bytecodes::_f2d:
1.1403 + __ cvtss2sd(xmm0, xmm0);
1.1404 + break;
1.1405 + case Bytecodes::_d2i:
1.1406 + {
1.1407 + Label L;
1.1408 + __ cvttsd2sil(rax, xmm0);
1.1409 + __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1.1410 + __ jcc(Assembler::notEqual, L);
1.1411 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1.1412 + __ bind(L);
1.1413 + }
1.1414 + break;
1.1415 + case Bytecodes::_d2l:
1.1416 + {
1.1417 + Label L;
1.1418 + __ cvttsd2siq(rax, xmm0);
1.1419 + // NaN or overflow/underflow?
1.1420 + __ cmp64(rax, ExternalAddress((address) &is_nan));
1.1421 + __ jcc(Assembler::notEqual, L);
1.1422 + __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1.1423 + __ bind(L);
1.1424 + }
1.1425 + break;
1.1426 + case Bytecodes::_d2f:
1.1427 + __ cvtsd2ss(xmm0, xmm0);
1.1428 + break;
1.1429 + default:
1.1430 + ShouldNotReachHere();
1.1431 + }
1.1432 +}
1.1433 +
1.1434 +void TemplateTable::lcmp() {
1.1435 + transition(ltos, itos);
1.1436 + Label done;
1.1437 + __ pop_l(rdx);
1.1438 + __ cmpq(rdx, rax);
1.1439 + __ movl(rax, -1);
1.1440 + __ jccb(Assembler::less, done);
1.1441 + __ setb(Assembler::notEqual, rax);
1.1442 + __ movzbl(rax, rax);
1.1443 + __ bind(done);
1.1444 +}
1.1445 +
1.1446 +void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1.1447 + Label done;
1.1448 + if (is_float) {
1.1449 + // XXX get rid of pop here, use ... reg, mem32
1.1450 + __ pop_f(xmm1);
1.1451 + __ ucomiss(xmm1, xmm0);
1.1452 + } else {
1.1453 + // XXX get rid of pop here, use ... reg, mem64
1.1454 + __ pop_d(xmm1);
1.1455 + __ ucomisd(xmm1, xmm0);
1.1456 + }
1.1457 + if (unordered_result < 0) {
1.1458 + __ movl(rax, -1);
1.1459 + __ jccb(Assembler::parity, done);
1.1460 + __ jccb(Assembler::below, done);
1.1461 + __ setb(Assembler::notEqual, rdx);
1.1462 + __ movzbl(rax, rdx);
1.1463 + } else {
1.1464 + __ movl(rax, 1);
1.1465 + __ jccb(Assembler::parity, done);
1.1466 + __ jccb(Assembler::above, done);
1.1467 + __ movl(rax, 0);
1.1468 + __ jccb(Assembler::equal, done);
1.1469 + __ decrementl(rax);
1.1470 + }
1.1471 + __ bind(done);
1.1472 +}
1.1473 +
1.1474 +void TemplateTable::branch(bool is_jsr, bool is_wide) {
1.1475 + __ get_method(rcx); // rcx holds method
1.1476 + __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
1.1477 + // holds bumped taken count
1.1478 +
1.1479 + const ByteSize be_offset = methodOopDesc::backedge_counter_offset() +
1.1480 + InvocationCounter::counter_offset();
1.1481 + const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() +
1.1482 + InvocationCounter::counter_offset();
1.1483 + const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1.1484 +
1.1485 + // Load up edx with the branch displacement
1.1486 + __ movl(rdx, at_bcp(1));
1.1487 + __ bswapl(rdx);
1.1488 +
1.1489 + if (!is_wide) {
1.1490 + __ sarl(rdx, 16);
1.1491 + }
1.1492 + __ movslq(rdx, rdx);
1.1493 +
1.1494 + // Handle all the JSR stuff here, then exit.
1.1495 + // It's much shorter and cleaner than intermingling with the non-JSR
1.1496 + // normal-branch stuff occuring below.
1.1497 + if (is_jsr) {
1.1498 + // Pre-load the next target bytecode into rbx
1.1499 + __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1, 0));
1.1500 +
1.1501 + // compute return address as bci in rax
1.1502 + __ leaq(rax, at_bcp((is_wide ? 5 : 3) -
1.1503 + in_bytes(constMethodOopDesc::codes_offset())));
1.1504 + __ subq(rax, Address(rcx, methodOopDesc::const_offset()));
1.1505 + // Adjust the bcp in r13 by the displacement in rdx
1.1506 + __ addq(r13, rdx);
1.1507 + // jsr returns atos that is not an oop
1.1508 + __ push_i(rax);
1.1509 + __ dispatch_only(vtos);
1.1510 + return;
1.1511 + }
1.1512 +
1.1513 + // Normal (non-jsr) branch handling
1.1514 +
1.1515 + // Adjust the bcp in r13 by the displacement in rdx
1.1516 + __ addq(r13, rdx);
1.1517 +
1.1518 + assert(UseLoopCounter || !UseOnStackReplacement,
1.1519 + "on-stack-replacement requires loop counters");
1.1520 + Label backedge_counter_overflow;
1.1521 + Label profile_method;
1.1522 + Label dispatch;
1.1523 + if (UseLoopCounter) {
1.1524 + // increment backedge counter for backward branches
1.1525 + // rax: MDO
1.1526 + // ebx: MDO bumped taken-count
1.1527 + // rcx: method
1.1528 + // rdx: target offset
1.1529 + // r13: target bcp
1.1530 + // r14: locals pointer
1.1531 + __ testl(rdx, rdx); // check if forward or backward branch
1.1532 + __ jcc(Assembler::positive, dispatch); // count only if backward branch
1.1533 +
1.1534 + // increment counter
1.1535 + __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1.1536 + __ incrementl(rax, InvocationCounter::count_increment); // increment
1.1537 + // counter
1.1538 + __ movl(Address(rcx, be_offset), rax); // store counter
1.1539 +
1.1540 + __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1.1541 + __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1.1542 + __ addl(rax, Address(rcx, be_offset)); // add both counters
1.1543 +
1.1544 + if (ProfileInterpreter) {
1.1545 + // Test to see if we should create a method data oop
1.1546 + __ cmp32(rax,
1.1547 + ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1.1548 + __ jcc(Assembler::less, dispatch);
1.1549 +
1.1550 + // if no method data exists, go to profile method
1.1551 + __ test_method_data_pointer(rax, profile_method);
1.1552 +
1.1553 + if (UseOnStackReplacement) {
1.1554 + // check for overflow against ebx which is the MDO taken count
1.1555 + __ cmp32(rbx,
1.1556 + ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1.1557 + __ jcc(Assembler::below, dispatch);
1.1558 +
1.1559 + // When ProfileInterpreter is on, the backedge_count comes
1.1560 + // from the methodDataOop, which value does not get reset on
1.1561 + // the call to frequency_counter_overflow(). To avoid
1.1562 + // excessive calls to the overflow routine while the method is
1.1563 + // being compiled, add a second test to make sure the overflow
1.1564 + // function is called only once every overflow_frequency.
1.1565 + const int overflow_frequency = 1024;
1.1566 + __ andl(rbx, overflow_frequency - 1);
1.1567 + __ jcc(Assembler::zero, backedge_counter_overflow);
1.1568 +
1.1569 + }
1.1570 + } else {
1.1571 + if (UseOnStackReplacement) {
1.1572 + // check for overflow against eax, which is the sum of the
1.1573 + // counters
1.1574 + __ cmp32(rax,
1.1575 + ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1.1576 + __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1.1577 +
1.1578 + }
1.1579 + }
1.1580 + __ bind(dispatch);
1.1581 + }
1.1582 +
1.1583 + // Pre-load the next target bytecode into rbx
1.1584 + __ load_unsigned_byte(rbx, Address(r13, 0));
1.1585 +
1.1586 + // continue with the bytecode @ target
1.1587 + // eax: return bci for jsr's, unused otherwise
1.1588 + // ebx: target bytecode
1.1589 + // r13: target bcp
1.1590 + __ dispatch_only(vtos);
1.1591 +
1.1592 + if (UseLoopCounter) {
1.1593 + if (ProfileInterpreter) {
1.1594 + // Out-of-line code to allocate method data oop.
1.1595 + __ bind(profile_method);
1.1596 + __ call_VM(noreg,
1.1597 + CAST_FROM_FN_PTR(address,
1.1598 + InterpreterRuntime::profile_method), r13);
1.1599 + __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1.1600 + __ movq(rcx, Address(rbp, method_offset));
1.1601 + __ movq(rcx, Address(rcx,
1.1602 + in_bytes(methodOopDesc::method_data_offset())));
1.1603 + __ movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
1.1604 + rcx);
1.1605 + __ test_method_data_pointer(rcx, dispatch);
1.1606 + // offset non-null mdp by MDO::data_offset() + IR::profile_method()
1.1607 + __ addq(rcx, in_bytes(methodDataOopDesc::data_offset()));
1.1608 + __ addq(rcx, rax);
1.1609 + __ movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
1.1610 + rcx);
1.1611 + __ jmp(dispatch);
1.1612 + }
1.1613 +
1.1614 + if (UseOnStackReplacement) {
1.1615 + // invocation counter overflow
1.1616 + __ bind(backedge_counter_overflow);
1.1617 + __ negq(rdx);
1.1618 + __ addq(rdx, r13); // branch bcp
1.1619 + // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1.1620 + __ call_VM(noreg,
1.1621 + CAST_FROM_FN_PTR(address,
1.1622 + InterpreterRuntime::frequency_counter_overflow),
1.1623 + rdx);
1.1624 + __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1.1625 +
1.1626 + // rax: osr nmethod (osr ok) or NULL (osr not possible)
1.1627 + // ebx: target bytecode
1.1628 + // rdx: scratch
1.1629 + // r14: locals pointer
1.1630 + // r13: bcp
1.1631 + __ testq(rax, rax); // test result
1.1632 + __ jcc(Assembler::zero, dispatch); // no osr if null
1.1633 + // nmethod may have been invalidated (VM may block upon call_VM return)
1.1634 + __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1.1635 + __ cmpl(rcx, InvalidOSREntryBci);
1.1636 + __ jcc(Assembler::equal, dispatch);
1.1637 +
1.1638 + // We have the address of an on stack replacement routine in eax
1.1639 + // We need to prepare to execute the OSR method. First we must
1.1640 + // migrate the locals and monitors off of the stack.
1.1641 +
1.1642 + __ movq(r13, rax); // save the nmethod
1.1643 +
1.1644 + call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1.1645 +
1.1646 + // eax is OSR buffer, move it to expected parameter location
1.1647 + __ movq(j_rarg0, rax);
1.1648 +
1.1649 + // We use j_rarg definitions here so that registers don't conflict as parameter
1.1650 + // registers change across platforms as we are in the midst of a calling
1.1651 + // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
1.1652 +
1.1653 + const Register retaddr = j_rarg2;
1.1654 + const Register sender_sp = j_rarg1;
1.1655 +
1.1656 + // pop the interpreter frame
1.1657 + __ movq(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1.1658 + __ leave(); // remove frame anchor
1.1659 + __ popq(retaddr); // get return address
1.1660 + __ movq(rsp, sender_sp); // set sp to sender sp
1.1661 + // Ensure compiled code always sees stack at proper alignment
1.1662 + __ andq(rsp, -(StackAlignmentInBytes));
1.1663 +
1.1664 + // unlike x86 we need no specialized return from compiled code
1.1665 + // to the interpreter or the call stub.
1.1666 +
1.1667 + // push the return address
1.1668 + __ pushq(retaddr);
1.1669 +
1.1670 + // and begin the OSR nmethod
1.1671 + __ jmp(Address(r13, nmethod::osr_entry_point_offset()));
1.1672 + }
1.1673 + }
1.1674 +}
1.1675 +
1.1676 +
1.1677 +void TemplateTable::if_0cmp(Condition cc) {
1.1678 + transition(itos, vtos);
1.1679 + // assume branch is more often taken than not (loops use backward branches)
1.1680 + Label not_taken;
1.1681 + __ testl(rax, rax);
1.1682 + __ jcc(j_not(cc), not_taken);
1.1683 + branch(false, false);
1.1684 + __ bind(not_taken);
1.1685 + __ profile_not_taken_branch(rax);
1.1686 +}
1.1687 +
1.1688 +void TemplateTable::if_icmp(Condition cc) {
1.1689 + transition(itos, vtos);
1.1690 + // assume branch is more often taken than not (loops use backward branches)
1.1691 + Label not_taken;
1.1692 + __ pop_i(rdx);
1.1693 + __ cmpl(rdx, rax);
1.1694 + __ jcc(j_not(cc), not_taken);
1.1695 + branch(false, false);
1.1696 + __ bind(not_taken);
1.1697 + __ profile_not_taken_branch(rax);
1.1698 +}
1.1699 +
1.1700 +void TemplateTable::if_nullcmp(Condition cc) {
1.1701 + transition(atos, vtos);
1.1702 + // assume branch is more often taken than not (loops use backward branches)
1.1703 + Label not_taken;
1.1704 + __ testq(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 +void TemplateTable::if_acmp(Condition cc) {
1.1712 + transition(atos, vtos);
1.1713 + // assume branch is more often taken than not (loops use backward branches)
1.1714 + Label not_taken;
1.1715 + __ pop_ptr(rdx);
1.1716 + __ cmpq(rdx, rax);
1.1717 + __ jcc(j_not(cc), not_taken);
1.1718 + branch(false, false);
1.1719 + __ bind(not_taken);
1.1720 + __ profile_not_taken_branch(rax);
1.1721 +}
1.1722 +
1.1723 +void TemplateTable::ret() {
1.1724 + transition(vtos, vtos);
1.1725 + locals_index(rbx);
1.1726 + __ movq(rbx, aaddress(rbx)); // get return bci, compute return bcp
1.1727 + __ profile_ret(rbx, rcx);
1.1728 + __ get_method(rax);
1.1729 + __ movq(r13, Address(rax, methodOopDesc::const_offset()));
1.1730 + __ leaq(r13, Address(r13, rbx, Address::times_1,
1.1731 + constMethodOopDesc::codes_offset()));
1.1732 + __ dispatch_next(vtos);
1.1733 +}
1.1734 +
1.1735 +void TemplateTable::wide_ret() {
1.1736 + transition(vtos, vtos);
1.1737 + locals_index_wide(rbx);
1.1738 + __ movq(rbx, aaddress(rbx)); // get return bci, compute return bcp
1.1739 + __ profile_ret(rbx, rcx);
1.1740 + __ get_method(rax);
1.1741 + __ movq(r13, Address(rax, methodOopDesc::const_offset()));
1.1742 + __ leaq(r13, Address(r13, rbx, Address::times_1, constMethodOopDesc::codes_offset()));
1.1743 + __ dispatch_next(vtos);
1.1744 +}
1.1745 +
1.1746 +void TemplateTable::tableswitch() {
1.1747 + Label default_case, continue_execution;
1.1748 + transition(itos, vtos);
1.1749 + // align r13
1.1750 + __ leaq(rbx, at_bcp(BytesPerInt));
1.1751 + __ andq(rbx, -BytesPerInt);
1.1752 + // load lo & hi
1.1753 + __ movl(rcx, Address(rbx, BytesPerInt));
1.1754 + __ movl(rdx, Address(rbx, 2 * BytesPerInt));
1.1755 + __ bswapl(rcx);
1.1756 + __ bswapl(rdx);
1.1757 + // check against lo & hi
1.1758 + __ cmpl(rax, rcx);
1.1759 + __ jcc(Assembler::less, default_case);
1.1760 + __ cmpl(rax, rdx);
1.1761 + __ jcc(Assembler::greater, default_case);
1.1762 + // lookup dispatch offset
1.1763 + __ subl(rax, rcx);
1.1764 + __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1.1765 + __ profile_switch_case(rax, rbx, rcx);
1.1766 + // continue execution
1.1767 + __ bind(continue_execution);
1.1768 + __ bswapl(rdx);
1.1769 + __ movslq(rdx, rdx);
1.1770 + __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1.1771 + __ addq(r13, rdx);
1.1772 + __ dispatch_only(vtos);
1.1773 + // handle default
1.1774 + __ bind(default_case);
1.1775 + __ profile_switch_default(rax);
1.1776 + __ movl(rdx, Address(rbx, 0));
1.1777 + __ jmp(continue_execution);
1.1778 +}
1.1779 +
1.1780 +void TemplateTable::lookupswitch() {
1.1781 + transition(itos, itos);
1.1782 + __ stop("lookupswitch bytecode should have been rewritten");
1.1783 +}
1.1784 +
1.1785 +void TemplateTable::fast_linearswitch() {
1.1786 + transition(itos, vtos);
1.1787 + Label loop_entry, loop, found, continue_execution;
1.1788 + // bswap rax so we can avoid bswapping the table entries
1.1789 + __ bswapl(rax);
1.1790 + // align r13
1.1791 + __ leaq(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
1.1792 + // this instruction (change offsets
1.1793 + // below)
1.1794 + __ andq(rbx, -BytesPerInt);
1.1795 + // set counter
1.1796 + __ movl(rcx, Address(rbx, BytesPerInt));
1.1797 + __ bswapl(rcx);
1.1798 + __ jmpb(loop_entry);
1.1799 + // table search
1.1800 + __ bind(loop);
1.1801 + __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
1.1802 + __ jcc(Assembler::equal, found);
1.1803 + __ bind(loop_entry);
1.1804 + __ decrementl(rcx);
1.1805 + __ jcc(Assembler::greaterEqual, loop);
1.1806 + // default case
1.1807 + __ profile_switch_default(rax);
1.1808 + __ movl(rdx, Address(rbx, 0));
1.1809 + __ jmp(continue_execution);
1.1810 + // entry found -> get offset
1.1811 + __ bind(found);
1.1812 + __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
1.1813 + __ profile_switch_case(rcx, rax, rbx);
1.1814 + // continue execution
1.1815 + __ bind(continue_execution);
1.1816 + __ bswapl(rdx);
1.1817 + __ movslq(rdx, rdx);
1.1818 + __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1.1819 + __ addq(r13, rdx);
1.1820 + __ dispatch_only(vtos);
1.1821 +}
1.1822 +
1.1823 +void TemplateTable::fast_binaryswitch() {
1.1824 + transition(itos, vtos);
1.1825 + // Implementation using the following core algorithm:
1.1826 + //
1.1827 + // int binary_search(int key, LookupswitchPair* array, int n) {
1.1828 + // // Binary search according to "Methodik des Programmierens" by
1.1829 + // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1.1830 + // int i = 0;
1.1831 + // int j = n;
1.1832 + // while (i+1 < j) {
1.1833 + // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1.1834 + // // with Q: for all i: 0 <= i < n: key < a[i]
1.1835 + // // where a stands for the array and assuming that the (inexisting)
1.1836 + // // element a[n] is infinitely big.
1.1837 + // int h = (i + j) >> 1;
1.1838 + // // i < h < j
1.1839 + // if (key < array[h].fast_match()) {
1.1840 + // j = h;
1.1841 + // } else {
1.1842 + // i = h;
1.1843 + // }
1.1844 + // }
1.1845 + // // R: a[i] <= key < a[i+1] or Q
1.1846 + // // (i.e., if key is within array, i is the correct index)
1.1847 + // return i;
1.1848 + // }
1.1849 +
1.1850 + // Register allocation
1.1851 + const Register key = rax; // already set (tosca)
1.1852 + const Register array = rbx;
1.1853 + const Register i = rcx;
1.1854 + const Register j = rdx;
1.1855 + const Register h = rdi;
1.1856 + const Register temp = rsi;
1.1857 +
1.1858 + // Find array start
1.1859 + __ leaq(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
1.1860 + // get rid of this
1.1861 + // instruction (change
1.1862 + // offsets below)
1.1863 + __ andq(array, -BytesPerInt);
1.1864 +
1.1865 + // Initialize i & j
1.1866 + __ xorl(i, i); // i = 0;
1.1867 + __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
1.1868 +
1.1869 + // Convert j into native byteordering
1.1870 + __ bswapl(j);
1.1871 +
1.1872 + // And start
1.1873 + Label entry;
1.1874 + __ jmp(entry);
1.1875 +
1.1876 + // binary search loop
1.1877 + {
1.1878 + Label loop;
1.1879 + __ bind(loop);
1.1880 + // int h = (i + j) >> 1;
1.1881 + __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1.1882 + __ sarl(h, 1); // h = (i + j) >> 1;
1.1883 + // if (key < array[h].fast_match()) {
1.1884 + // j = h;
1.1885 + // } else {
1.1886 + // i = h;
1.1887 + // }
1.1888 + // Convert array[h].match to native byte-ordering before compare
1.1889 + __ movl(temp, Address(array, h, Address::times_8));
1.1890 + __ bswapl(temp);
1.1891 + __ cmpl(key, temp);
1.1892 + // j = h if (key < array[h].fast_match())
1.1893 + __ cmovl(Assembler::less, j, h);
1.1894 + // i = h if (key >= array[h].fast_match())
1.1895 + __ cmovl(Assembler::greaterEqual, i, h);
1.1896 + // while (i+1 < j)
1.1897 + __ bind(entry);
1.1898 + __ leal(h, Address(i, 1)); // i+1
1.1899 + __ cmpl(h, j); // i+1 < j
1.1900 + __ jcc(Assembler::less, loop);
1.1901 + }
1.1902 +
1.1903 + // end of binary search, result index is i (must check again!)
1.1904 + Label default_case;
1.1905 + // Convert array[i].match to native byte-ordering before compare
1.1906 + __ movl(temp, Address(array, i, Address::times_8));
1.1907 + __ bswapl(temp);
1.1908 + __ cmpl(key, temp);
1.1909 + __ jcc(Assembler::notEqual, default_case);
1.1910 +
1.1911 + // entry found -> j = offset
1.1912 + __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
1.1913 + __ profile_switch_case(i, key, array);
1.1914 + __ bswapl(j);
1.1915 + __ movslq(j, j);
1.1916 + __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
1.1917 + __ addq(r13, j);
1.1918 + __ dispatch_only(vtos);
1.1919 +
1.1920 + // default case -> j = default offset
1.1921 + __ bind(default_case);
1.1922 + __ profile_switch_default(i);
1.1923 + __ movl(j, Address(array, -2 * BytesPerInt));
1.1924 + __ bswapl(j);
1.1925 + __ movslq(j, j);
1.1926 + __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
1.1927 + __ addq(r13, j);
1.1928 + __ dispatch_only(vtos);
1.1929 +}
1.1930 +
1.1931 +
1.1932 +void TemplateTable::_return(TosState state) {
1.1933 + transition(state, state);
1.1934 + assert(_desc->calls_vm(),
1.1935 + "inconsistent calls_vm information"); // call in remove_activation
1.1936 +
1.1937 + if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
1.1938 + assert(state == vtos, "only valid state");
1.1939 + __ movq(c_rarg1, aaddress(0));
1.1940 + __ movq(rdi, Address(c_rarg1, oopDesc::klass_offset_in_bytes()));
1.1941 + __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
1.1942 + __ testl(rdi, JVM_ACC_HAS_FINALIZER);
1.1943 + Label skip_register_finalizer;
1.1944 + __ jcc(Assembler::zero, skip_register_finalizer);
1.1945 +
1.1946 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
1.1947 +
1.1948 + __ bind(skip_register_finalizer);
1.1949 + }
1.1950 +
1.1951 + __ remove_activation(state, r13);
1.1952 + __ jmp(r13);
1.1953 +}
1.1954 +
1.1955 +// ----------------------------------------------------------------------------
1.1956 +// Volatile variables demand their effects be made known to all CPU's
1.1957 +// in order. Store buffers on most chips allow reads & writes to
1.1958 +// reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
1.1959 +// without some kind of memory barrier (i.e., it's not sufficient that
1.1960 +// the interpreter does not reorder volatile references, the hardware
1.1961 +// also must not reorder them).
1.1962 +//
1.1963 +// According to the new Java Memory Model (JMM):
1.1964 +// (1) All volatiles are serialized wrt to each other. ALSO reads &
1.1965 +// writes act as aquire & release, so:
1.1966 +// (2) A read cannot let unrelated NON-volatile memory refs that
1.1967 +// happen after the read float up to before the read. It's OK for
1.1968 +// non-volatile memory refs that happen before the volatile read to
1.1969 +// float down below it.
1.1970 +// (3) Similar a volatile write cannot let unrelated NON-volatile
1.1971 +// memory refs that happen BEFORE the write float down to after the
1.1972 +// write. It's OK for non-volatile memory refs that happen after the
1.1973 +// volatile write to float up before it.
1.1974 +//
1.1975 +// We only put in barriers around volatile refs (they are expensive),
1.1976 +// not _between_ memory refs (that would require us to track the
1.1977 +// flavor of the previous memory refs). Requirements (2) and (3)
1.1978 +// require some barriers before volatile stores and after volatile
1.1979 +// loads. These nearly cover requirement (1) but miss the
1.1980 +// volatile-store-volatile-load case. This final case is placed after
1.1981 +// volatile-stores although it could just as well go before
1.1982 +// volatile-loads.
1.1983 +void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits
1.1984 + order_constraint) {
1.1985 + // Helper function to insert a is-volatile test and memory barrier
1.1986 + if (os::is_MP()) { // Not needed on single CPU
1.1987 + __ membar(order_constraint);
1.1988 + }
1.1989 +}
1.1990 +
1.1991 +void TemplateTable::resolve_cache_and_index(int byte_no,
1.1992 + Register Rcache,
1.1993 + Register index) {
1.1994 + assert(byte_no == 1 || byte_no == 2, "byte_no out of range");
1.1995 +
1.1996 + const Register temp = rbx;
1.1997 + assert_different_registers(Rcache, index, temp);
1.1998 +
1.1999 + const int shift_count = (1 + byte_no) * BitsPerByte;
1.2000 + Label resolved;
1.2001 + __ get_cache_and_index_at_bcp(Rcache, index, 1);
1.2002 + __ movl(temp, Address(Rcache,
1.2003 + index, Address::times_8,
1.2004 + constantPoolCacheOopDesc::base_offset() +
1.2005 + ConstantPoolCacheEntry::indices_offset()));
1.2006 + __ shrl(temp, shift_count);
1.2007 + // have we resolved this bytecode?
1.2008 + __ andl(temp, 0xFF);
1.2009 + __ cmpl(temp, (int) bytecode());
1.2010 + __ jcc(Assembler::equal, resolved);
1.2011 +
1.2012 + // resolve first time through
1.2013 + address entry;
1.2014 + switch (bytecode()) {
1.2015 + case Bytecodes::_getstatic:
1.2016 + case Bytecodes::_putstatic:
1.2017 + case Bytecodes::_getfield:
1.2018 + case Bytecodes::_putfield:
1.2019 + entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);
1.2020 + break;
1.2021 + case Bytecodes::_invokevirtual:
1.2022 + case Bytecodes::_invokespecial:
1.2023 + case Bytecodes::_invokestatic:
1.2024 + case Bytecodes::_invokeinterface:
1.2025 + entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
1.2026 + break;
1.2027 + default:
1.2028 + ShouldNotReachHere();
1.2029 + break;
1.2030 + }
1.2031 + __ movl(temp, (int) bytecode());
1.2032 + __ call_VM(noreg, entry, temp);
1.2033 +
1.2034 + // Update registers with resolved info
1.2035 + __ get_cache_and_index_at_bcp(Rcache, index, 1);
1.2036 + __ bind(resolved);
1.2037 +}
1.2038 +
1.2039 +// The Rcache and index registers must be set before call
1.2040 +void TemplateTable::load_field_cp_cache_entry(Register obj,
1.2041 + Register cache,
1.2042 + Register index,
1.2043 + Register off,
1.2044 + Register flags,
1.2045 + bool is_static = false) {
1.2046 + assert_different_registers(cache, index, flags, off);
1.2047 +
1.2048 + ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
1.2049 + // Field offset
1.2050 + __ movq(off, Address(cache, index, Address::times_8,
1.2051 + in_bytes(cp_base_offset +
1.2052 + ConstantPoolCacheEntry::f2_offset())));
1.2053 + // Flags
1.2054 + __ movl(flags, Address(cache, index, Address::times_8,
1.2055 + in_bytes(cp_base_offset +
1.2056 + ConstantPoolCacheEntry::flags_offset())));
1.2057 +
1.2058 + // klass overwrite register
1.2059 + if (is_static) {
1.2060 + __ movq(obj, Address(cache, index, Address::times_8,
1.2061 + in_bytes(cp_base_offset +
1.2062 + ConstantPoolCacheEntry::f1_offset())));
1.2063 + }
1.2064 +}
1.2065 +
1.2066 +void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
1.2067 + Register method,
1.2068 + Register itable_index,
1.2069 + Register flags,
1.2070 + bool is_invokevirtual,
1.2071 + bool is_invokevfinal /*unused*/) {
1.2072 + // setup registers
1.2073 + const Register cache = rcx;
1.2074 + const Register index = rdx;
1.2075 + assert_different_registers(method, flags);
1.2076 + assert_different_registers(method, cache, index);
1.2077 + assert_different_registers(itable_index, flags);
1.2078 + assert_different_registers(itable_index, cache, index);
1.2079 + // determine constant pool cache field offsets
1.2080 + const int method_offset = in_bytes(
1.2081 + constantPoolCacheOopDesc::base_offset() +
1.2082 + (is_invokevirtual
1.2083 + ? ConstantPoolCacheEntry::f2_offset()
1.2084 + : ConstantPoolCacheEntry::f1_offset()));
1.2085 + const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
1.2086 + ConstantPoolCacheEntry::flags_offset());
1.2087 + // access constant pool cache fields
1.2088 + const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
1.2089 + ConstantPoolCacheEntry::f2_offset());
1.2090 +
1.2091 + resolve_cache_and_index(byte_no, cache, index);
1.2092 +
1.2093 + assert(wordSize == 8, "adjust code below");
1.2094 + __ movq(method, Address(cache, index, Address::times_8, method_offset));
1.2095 + if (itable_index != noreg) {
1.2096 + __ movq(itable_index,
1.2097 + Address(cache, index, Address::times_8, index_offset));
1.2098 + }
1.2099 + __ movl(flags , Address(cache, index, Address::times_8, flags_offset));
1.2100 +}
1.2101 +
1.2102 +
1.2103 +// The registers cache and index expected to be set before call.
1.2104 +// Correct values of the cache and index registers are preserved.
1.2105 +void TemplateTable::jvmti_post_field_access(Register cache, Register index,
1.2106 + bool is_static, bool has_tos) {
1.2107 + // do the JVMTI work here to avoid disturbing the register state below
1.2108 + // We use c_rarg registers here because we want to use the register used in
1.2109 + // the call to the VM
1.2110 + if (JvmtiExport::can_post_field_access()) {
1.2111 + // Check to see if a field access watch has been set before we
1.2112 + // take the time to call into the VM.
1.2113 + Label L1;
1.2114 + assert_different_registers(cache, index, rax);
1.2115 + __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
1.2116 + __ testl(rax, rax);
1.2117 + __ jcc(Assembler::zero, L1);
1.2118 +
1.2119 + __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
1.2120 +
1.2121 + // cache entry pointer
1.2122 + __ addq(c_rarg2, in_bytes(constantPoolCacheOopDesc::base_offset()));
1.2123 + __ shll(c_rarg3, LogBytesPerWord);
1.2124 + __ addq(c_rarg2, c_rarg3);
1.2125 + if (is_static) {
1.2126 + __ xorl(c_rarg1, c_rarg1); // NULL object reference
1.2127 + } else {
1.2128 + __ movq(c_rarg1, at_tos()); // get object pointer without popping it
1.2129 + __ verify_oop(c_rarg1);
1.2130 + }
1.2131 + // c_rarg1: object pointer or NULL
1.2132 + // c_rarg2: cache entry pointer
1.2133 + // c_rarg3: jvalue object on the stack
1.2134 + __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1.2135 + InterpreterRuntime::post_field_access),
1.2136 + c_rarg1, c_rarg2, c_rarg3);
1.2137 + __ get_cache_and_index_at_bcp(cache, index, 1);
1.2138 + __ bind(L1);
1.2139 + }
1.2140 +}
1.2141 +
1.2142 +void TemplateTable::pop_and_check_object(Register r) {
1.2143 + __ pop_ptr(r);
1.2144 + __ null_check(r); // for field access must check obj.
1.2145 + __ verify_oop(r);
1.2146 +}
1.2147 +
1.2148 +void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
1.2149 + transition(vtos, vtos);
1.2150 +
1.2151 + const Register cache = rcx;
1.2152 + const Register index = rdx;
1.2153 + const Register obj = c_rarg3;
1.2154 + const Register off = rbx;
1.2155 + const Register flags = rax;
1.2156 + const Register bc = c_rarg3; // uses same reg as obj, so don't mix them
1.2157 +
1.2158 + resolve_cache_and_index(byte_no, cache, index);
1.2159 + jvmti_post_field_access(cache, index, is_static, false);
1.2160 + load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
1.2161 +
1.2162 + if (!is_static) {
1.2163 + // obj is on the stack
1.2164 + pop_and_check_object(obj);
1.2165 + }
1.2166 +
1.2167 + const Address field(obj, off, Address::times_1);
1.2168 +
1.2169 + Label Done, notByte, notInt, notShort, notChar,
1.2170 + notLong, notFloat, notObj, notDouble;
1.2171 +
1.2172 + __ shrl(flags, ConstantPoolCacheEntry::tosBits);
1.2173 + assert(btos == 0, "change code, btos != 0");
1.2174 +
1.2175 + __ andl(flags, 0x0F);
1.2176 + __ jcc(Assembler::notZero, notByte);
1.2177 + // btos
1.2178 + __ load_signed_byte(rax, field);
1.2179 + __ push(btos);
1.2180 + // Rewrite bytecode to be faster
1.2181 + if (!is_static) {
1.2182 + patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
1.2183 + }
1.2184 + __ jmp(Done);
1.2185 +
1.2186 + __ bind(notByte);
1.2187 + __ cmpl(flags, atos);
1.2188 + __ jcc(Assembler::notEqual, notObj);
1.2189 + // atos
1.2190 + __ movq(rax, field);
1.2191 + __ push(atos);
1.2192 + if (!is_static) {
1.2193 + patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
1.2194 + }
1.2195 + __ jmp(Done);
1.2196 +
1.2197 + __ bind(notObj);
1.2198 + __ cmpl(flags, itos);
1.2199 + __ jcc(Assembler::notEqual, notInt);
1.2200 + // itos
1.2201 + __ movl(rax, field);
1.2202 + __ push(itos);
1.2203 + // Rewrite bytecode to be faster
1.2204 + if (!is_static) {
1.2205 + patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
1.2206 + }
1.2207 + __ jmp(Done);
1.2208 +
1.2209 + __ bind(notInt);
1.2210 + __ cmpl(flags, ctos);
1.2211 + __ jcc(Assembler::notEqual, notChar);
1.2212 + // ctos
1.2213 + __ load_unsigned_word(rax, field);
1.2214 + __ push(ctos);
1.2215 + // Rewrite bytecode to be faster
1.2216 + if (!is_static) {
1.2217 + patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
1.2218 + }
1.2219 + __ jmp(Done);
1.2220 +
1.2221 + __ bind(notChar);
1.2222 + __ cmpl(flags, stos);
1.2223 + __ jcc(Assembler::notEqual, notShort);
1.2224 + // stos
1.2225 + __ load_signed_word(rax, field);
1.2226 + __ push(stos);
1.2227 + // Rewrite bytecode to be faster
1.2228 + if (!is_static) {
1.2229 + patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
1.2230 + }
1.2231 + __ jmp(Done);
1.2232 +
1.2233 + __ bind(notShort);
1.2234 + __ cmpl(flags, ltos);
1.2235 + __ jcc(Assembler::notEqual, notLong);
1.2236 + // ltos
1.2237 + __ movq(rax, field);
1.2238 + __ push(ltos);
1.2239 + // Rewrite bytecode to be faster
1.2240 + if (!is_static) {
1.2241 + patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx);
1.2242 + }
1.2243 + __ jmp(Done);
1.2244 +
1.2245 + __ bind(notLong);
1.2246 + __ cmpl(flags, ftos);
1.2247 + __ jcc(Assembler::notEqual, notFloat);
1.2248 + // ftos
1.2249 + __ movflt(xmm0, field);
1.2250 + __ push(ftos);
1.2251 + // Rewrite bytecode to be faster
1.2252 + if (!is_static) {
1.2253 + patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
1.2254 + }
1.2255 + __ jmp(Done);
1.2256 +
1.2257 + __ bind(notFloat);
1.2258 +#ifdef ASSERT
1.2259 + __ cmpl(flags, dtos);
1.2260 + __ jcc(Assembler::notEqual, notDouble);
1.2261 +#endif
1.2262 + // dtos
1.2263 + __ movdbl(xmm0, field);
1.2264 + __ push(dtos);
1.2265 + // Rewrite bytecode to be faster
1.2266 + if (!is_static) {
1.2267 + patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
1.2268 + }
1.2269 +#ifdef ASSERT
1.2270 + __ jmp(Done);
1.2271 +
1.2272 + __ bind(notDouble);
1.2273 + __ stop("Bad state");
1.2274 +#endif
1.2275 +
1.2276 + __ bind(Done);
1.2277 + // [jk] not needed currently
1.2278 + // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
1.2279 + // Assembler::LoadStore));
1.2280 +}
1.2281 +
1.2282 +
1.2283 +void TemplateTable::getfield(int byte_no) {
1.2284 + getfield_or_static(byte_no, false);
1.2285 +}
1.2286 +
1.2287 +void TemplateTable::getstatic(int byte_no) {
1.2288 + getfield_or_static(byte_no, true);
1.2289 +}
1.2290 +
1.2291 +// The registers cache and index expected to be set before call.
1.2292 +// The function may destroy various registers, just not the cache and index registers.
1.2293 +void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
1.2294 + transition(vtos, vtos);
1.2295 +
1.2296 + ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
1.2297 +
1.2298 + if (JvmtiExport::can_post_field_modification()) {
1.2299 + // Check to see if a field modification watch has been set before
1.2300 + // we take the time to call into the VM.
1.2301 + Label L1;
1.2302 + assert_different_registers(cache, index, rax);
1.2303 + __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
1.2304 + __ testl(rax, rax);
1.2305 + __ jcc(Assembler::zero, L1);
1.2306 +
1.2307 + __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
1.2308 +
1.2309 + if (is_static) {
1.2310 + // Life is simple. Null out the object pointer.
1.2311 + __ xorl(c_rarg1, c_rarg1);
1.2312 + } else {
1.2313 + // Life is harder. The stack holds the value on top, followed by
1.2314 + // the object. We don't know the size of the value, though; it
1.2315 + // could be one or two words depending on its type. As a result,
1.2316 + // we must find the type to determine where the object is.
1.2317 + __ movl(c_rarg3, Address(c_rarg2, rscratch1,
1.2318 + Address::times_8,
1.2319 + in_bytes(cp_base_offset +
1.2320 + ConstantPoolCacheEntry::flags_offset())));
1.2321 + __ shrl(c_rarg3, ConstantPoolCacheEntry::tosBits);
1.2322 + // Make sure we don't need to mask rcx for tosBits after the
1.2323 + // above shift
1.2324 + ConstantPoolCacheEntry::verify_tosBits();
1.2325 + __ movq(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
1.2326 + __ cmpl(c_rarg3, ltos);
1.2327 + __ cmovq(Assembler::equal,
1.2328 + c_rarg1, at_tos_p2()); // ltos (two word jvalue)
1.2329 + __ cmpl(c_rarg3, dtos);
1.2330 + __ cmovq(Assembler::equal,
1.2331 + c_rarg1, at_tos_p2()); // dtos (two word jvalue)
1.2332 + }
1.2333 + // cache entry pointer
1.2334 + __ addq(c_rarg2, in_bytes(cp_base_offset));
1.2335 + __ shll(rscratch1, LogBytesPerWord);
1.2336 + __ addq(c_rarg2, rscratch1);
1.2337 + // object (tos)
1.2338 + __ movq(c_rarg3, rsp);
1.2339 + // c_rarg1: object pointer set up above (NULL if static)
1.2340 + // c_rarg2: cache entry pointer
1.2341 + // c_rarg3: jvalue object on the stack
1.2342 + __ call_VM(noreg,
1.2343 + CAST_FROM_FN_PTR(address,
1.2344 + InterpreterRuntime::post_field_modification),
1.2345 + c_rarg1, c_rarg2, c_rarg3);
1.2346 + __ get_cache_and_index_at_bcp(cache, index, 1);
1.2347 + __ bind(L1);
1.2348 + }
1.2349 +}
1.2350 +
1.2351 +void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
1.2352 + transition(vtos, vtos);
1.2353 +
1.2354 + const Register cache = rcx;
1.2355 + const Register index = rdx;
1.2356 + const Register obj = rcx;
1.2357 + const Register off = rbx;
1.2358 + const Register flags = rax;
1.2359 + const Register bc = c_rarg3;
1.2360 +
1.2361 + resolve_cache_and_index(byte_no, cache, index);
1.2362 + jvmti_post_field_mod(cache, index, is_static);
1.2363 + load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
1.2364 +
1.2365 + // [jk] not needed currently
1.2366 + // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
1.2367 + // Assembler::StoreStore));
1.2368 +
1.2369 + Label notVolatile, Done;
1.2370 + __ movl(rdx, flags);
1.2371 + __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
1.2372 + __ andl(rdx, 0x1);
1.2373 +
1.2374 + // field address
1.2375 + const Address field(obj, off, Address::times_1);
1.2376 +
1.2377 + Label notByte, notInt, notShort, notChar,
1.2378 + notLong, notFloat, notObj, notDouble;
1.2379 +
1.2380 + __ shrl(flags, ConstantPoolCacheEntry::tosBits);
1.2381 +
1.2382 + assert(btos == 0, "change code, btos != 0");
1.2383 + __ andl(flags, 0x0f);
1.2384 + __ jcc(Assembler::notZero, notByte);
1.2385 + // btos
1.2386 + __ pop(btos);
1.2387 + if (!is_static) pop_and_check_object(obj);
1.2388 + __ movb(field, rax);
1.2389 + if (!is_static) {
1.2390 + patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx);
1.2391 + }
1.2392 + __ jmp(Done);
1.2393 +
1.2394 + __ bind(notByte);
1.2395 + __ cmpl(flags, atos);
1.2396 + __ jcc(Assembler::notEqual, notObj);
1.2397 + // atos
1.2398 + __ pop(atos);
1.2399 + if (!is_static) pop_and_check_object(obj);
1.2400 + __ movq(field, rax);
1.2401 + __ store_check(obj, field); // Need to mark card
1.2402 + if (!is_static) {
1.2403 + patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx);
1.2404 + }
1.2405 + __ jmp(Done);
1.2406 +
1.2407 + __ bind(notObj);
1.2408 + __ cmpl(flags, itos);
1.2409 + __ jcc(Assembler::notEqual, notInt);
1.2410 + // itos
1.2411 + __ pop(itos);
1.2412 + if (!is_static) pop_and_check_object(obj);
1.2413 + __ movl(field, rax);
1.2414 + if (!is_static) {
1.2415 + patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx);
1.2416 + }
1.2417 + __ jmp(Done);
1.2418 +
1.2419 + __ bind(notInt);
1.2420 + __ cmpl(flags, ctos);
1.2421 + __ jcc(Assembler::notEqual, notChar);
1.2422 + // ctos
1.2423 + __ pop(ctos);
1.2424 + if (!is_static) pop_and_check_object(obj);
1.2425 + __ movw(field, rax);
1.2426 + if (!is_static) {
1.2427 + patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx);
1.2428 + }
1.2429 + __ jmp(Done);
1.2430 +
1.2431 + __ bind(notChar);
1.2432 + __ cmpl(flags, stos);
1.2433 + __ jcc(Assembler::notEqual, notShort);
1.2434 + // stos
1.2435 + __ pop(stos);
1.2436 + if (!is_static) pop_and_check_object(obj);
1.2437 + __ movw(field, rax);
1.2438 + if (!is_static) {
1.2439 + patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx);
1.2440 + }
1.2441 + __ jmp(Done);
1.2442 +
1.2443 + __ bind(notShort);
1.2444 + __ cmpl(flags, ltos);
1.2445 + __ jcc(Assembler::notEqual, notLong);
1.2446 + // ltos
1.2447 + __ pop(ltos);
1.2448 + if (!is_static) pop_and_check_object(obj);
1.2449 + __ movq(field, rax);
1.2450 + if (!is_static) {
1.2451 + patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx);
1.2452 + }
1.2453 + __ jmp(Done);
1.2454 +
1.2455 + __ bind(notLong);
1.2456 + __ cmpl(flags, ftos);
1.2457 + __ jcc(Assembler::notEqual, notFloat);
1.2458 + // ftos
1.2459 + __ pop(ftos);
1.2460 + if (!is_static) pop_and_check_object(obj);
1.2461 + __ movflt(field, xmm0);
1.2462 + if (!is_static) {
1.2463 + patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx);
1.2464 + }
1.2465 + __ jmp(Done);
1.2466 +
1.2467 + __ bind(notFloat);
1.2468 +#ifdef ASSERT
1.2469 + __ cmpl(flags, dtos);
1.2470 + __ jcc(Assembler::notEqual, notDouble);
1.2471 +#endif
1.2472 + // dtos
1.2473 + __ pop(dtos);
1.2474 + if (!is_static) pop_and_check_object(obj);
1.2475 + __ movdbl(field, xmm0);
1.2476 + if (!is_static) {
1.2477 + patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx);
1.2478 + }
1.2479 +
1.2480 +#ifdef ASSERT
1.2481 + __ jmp(Done);
1.2482 +
1.2483 + __ bind(notDouble);
1.2484 + __ stop("Bad state");
1.2485 +#endif
1.2486 +
1.2487 + __ bind(Done);
1.2488 + // Check for volatile store
1.2489 + __ testl(rdx, rdx);
1.2490 + __ jcc(Assembler::zero, notVolatile);
1.2491 + volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
1.2492 + Assembler::StoreStore));
1.2493 +
1.2494 + __ bind(notVolatile);
1.2495 +}
1.2496 +
1.2497 +void TemplateTable::putfield(int byte_no) {
1.2498 + putfield_or_static(byte_no, false);
1.2499 +}
1.2500 +
1.2501 +void TemplateTable::putstatic(int byte_no) {
1.2502 + putfield_or_static(byte_no, true);
1.2503 +}
1.2504 +
1.2505 +void TemplateTable::jvmti_post_fast_field_mod() {
1.2506 + if (JvmtiExport::can_post_field_modification()) {
1.2507 + // Check to see if a field modification watch has been set before
1.2508 + // we take the time to call into the VM.
1.2509 + Label L2;
1.2510 + __ mov32(c_rarg3, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
1.2511 + __ testl(c_rarg3, c_rarg3);
1.2512 + __ jcc(Assembler::zero, L2);
1.2513 + __ pop_ptr(rbx); // copy the object pointer from tos
1.2514 + __ verify_oop(rbx);
1.2515 + __ push_ptr(rbx); // put the object pointer back on tos
1.2516 + __ subq(rsp, sizeof(jvalue)); // add space for a jvalue object
1.2517 + __ movq(c_rarg3, rsp);
1.2518 + const Address field(c_rarg3, 0);
1.2519 +
1.2520 + switch (bytecode()) { // load values into the jvalue object
1.2521 + case Bytecodes::_fast_aputfield: // fall through
1.2522 + case Bytecodes::_fast_lputfield: __ movq(field, rax); break;
1.2523 + case Bytecodes::_fast_iputfield: __ movl(field, rax); break;
1.2524 + case Bytecodes::_fast_bputfield: __ movb(field, rax); break;
1.2525 + case Bytecodes::_fast_sputfield: // fall through
1.2526 + case Bytecodes::_fast_cputfield: __ movw(field, rax); break;
1.2527 + case Bytecodes::_fast_fputfield: __ movflt(field, xmm0); break;
1.2528 + case Bytecodes::_fast_dputfield: __ movdbl(field, xmm0); break;
1.2529 + default:
1.2530 + ShouldNotReachHere();
1.2531 + }
1.2532 +
1.2533 + // Save rax because call_VM() will clobber it, then use it for
1.2534 + // JVMTI purposes
1.2535 + __ pushq(rax);
1.2536 + // access constant pool cache entry
1.2537 + __ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1);
1.2538 + __ verify_oop(rbx);
1.2539 + // rbx: object pointer copied above
1.2540 + // c_rarg2: cache entry pointer
1.2541 + // c_rarg3: jvalue object on the stack
1.2542 + __ call_VM(noreg,
1.2543 + CAST_FROM_FN_PTR(address,
1.2544 + InterpreterRuntime::post_field_modification),
1.2545 + rbx, c_rarg2, c_rarg3);
1.2546 + __ popq(rax); // restore lower value
1.2547 + __ addq(rsp, sizeof(jvalue)); // release jvalue object space
1.2548 + __ bind(L2);
1.2549 + }
1.2550 +}
1.2551 +
1.2552 +void TemplateTable::fast_storefield(TosState state) {
1.2553 + transition(state, vtos);
1.2554 +
1.2555 + ByteSize base = constantPoolCacheOopDesc::base_offset();
1.2556 +
1.2557 + jvmti_post_fast_field_mod();
1.2558 +
1.2559 + // access constant pool cache
1.2560 + __ get_cache_and_index_at_bcp(rcx, rbx, 1);
1.2561 +
1.2562 + // test for volatile with rdx
1.2563 + __ movl(rdx, Address(rcx, rbx, Address::times_8,
1.2564 + in_bytes(base +
1.2565 + ConstantPoolCacheEntry::flags_offset())));
1.2566 +
1.2567 + // replace index with field offset from cache entry
1.2568 + __ movq(rbx, Address(rcx, rbx, Address::times_8,
1.2569 + in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
1.2570 +
1.2571 + // [jk] not needed currently
1.2572 + // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
1.2573 + // Assembler::StoreStore));
1.2574 +
1.2575 + Label notVolatile;
1.2576 + __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
1.2577 + __ andl(rdx, 0x1);
1.2578 +
1.2579 + // Get object from stack
1.2580 + pop_and_check_object(rcx);
1.2581 +
1.2582 + // field address
1.2583 + const Address field(rcx, rbx, Address::times_1);
1.2584 +
1.2585 + // access field
1.2586 + switch (bytecode()) {
1.2587 + case Bytecodes::_fast_aputfield:
1.2588 + __ movq(field, rax);
1.2589 + __ store_check(rcx, field);
1.2590 + break;
1.2591 + case Bytecodes::_fast_lputfield:
1.2592 + __ movq(field, rax);
1.2593 + break;
1.2594 + case Bytecodes::_fast_iputfield:
1.2595 + __ movl(field, rax);
1.2596 + break;
1.2597 + case Bytecodes::_fast_bputfield:
1.2598 + __ movb(field, rax);
1.2599 + break;
1.2600 + case Bytecodes::_fast_sputfield:
1.2601 + // fall through
1.2602 + case Bytecodes::_fast_cputfield:
1.2603 + __ movw(field, rax);
1.2604 + break;
1.2605 + case Bytecodes::_fast_fputfield:
1.2606 + __ movflt(field, xmm0);
1.2607 + break;
1.2608 + case Bytecodes::_fast_dputfield:
1.2609 + __ movdbl(field, xmm0);
1.2610 + break;
1.2611 + default:
1.2612 + ShouldNotReachHere();
1.2613 + }
1.2614 +
1.2615 + // Check for volatile store
1.2616 + __ testl(rdx, rdx);
1.2617 + __ jcc(Assembler::zero, notVolatile);
1.2618 + volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
1.2619 + Assembler::StoreStore));
1.2620 + __ bind(notVolatile);
1.2621 +}
1.2622 +
1.2623 +
1.2624 +void TemplateTable::fast_accessfield(TosState state) {
1.2625 + transition(atos, state);
1.2626 +
1.2627 + // Do the JVMTI work here to avoid disturbing the register state below
1.2628 + if (JvmtiExport::can_post_field_access()) {
1.2629 + // Check to see if a field access watch has been set before we
1.2630 + // take the time to call into the VM.
1.2631 + Label L1;
1.2632 + __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
1.2633 + __ testl(rcx, rcx);
1.2634 + __ jcc(Assembler::zero, L1);
1.2635 + // access constant pool cache entry
1.2636 + __ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1);
1.2637 + __ movq(r12, rax); // save object pointer before call_VM() clobbers it
1.2638 + __ verify_oop(rax);
1.2639 + __ movq(c_rarg1, rax);
1.2640 + // c_rarg1: object pointer copied above
1.2641 + // c_rarg2: cache entry pointer
1.2642 + __ call_VM(noreg,
1.2643 + CAST_FROM_FN_PTR(address,
1.2644 + InterpreterRuntime::post_field_access),
1.2645 + c_rarg1, c_rarg2);
1.2646 + __ movq(rax, r12); // restore object pointer
1.2647 + __ bind(L1);
1.2648 + }
1.2649 +
1.2650 + // access constant pool cache
1.2651 + __ get_cache_and_index_at_bcp(rcx, rbx, 1);
1.2652 + // replace index with field offset from cache entry
1.2653 + // [jk] not needed currently
1.2654 + // if (os::is_MP()) {
1.2655 + // __ movl(rdx, Address(rcx, rbx, Address::times_8,
1.2656 + // in_bytes(constantPoolCacheOopDesc::base_offset() +
1.2657 + // ConstantPoolCacheEntry::flags_offset())));
1.2658 + // __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
1.2659 + // __ andl(rdx, 0x1);
1.2660 + // }
1.2661 + __ movq(rbx, Address(rcx, rbx, Address::times_8,
1.2662 + in_bytes(constantPoolCacheOopDesc::base_offset() +
1.2663 + ConstantPoolCacheEntry::f2_offset())));
1.2664 +
1.2665 + // rax: object
1.2666 + __ verify_oop(rax);
1.2667 + __ null_check(rax);
1.2668 + Address field(rax, rbx, Address::times_1);
1.2669 +
1.2670 + // access field
1.2671 + switch (bytecode()) {
1.2672 + case Bytecodes::_fast_agetfield:
1.2673 + __ movq(rax, field);
1.2674 + __ verify_oop(rax);
1.2675 + break;
1.2676 + case Bytecodes::_fast_lgetfield:
1.2677 + __ movq(rax, field);
1.2678 + break;
1.2679 + case Bytecodes::_fast_igetfield:
1.2680 + __ movl(rax, field);
1.2681 + break;
1.2682 + case Bytecodes::_fast_bgetfield:
1.2683 + __ movsbl(rax, field);
1.2684 + break;
1.2685 + case Bytecodes::_fast_sgetfield:
1.2686 + __ load_signed_word(rax, field);
1.2687 + break;
1.2688 + case Bytecodes::_fast_cgetfield:
1.2689 + __ load_unsigned_word(rax, field);
1.2690 + break;
1.2691 + case Bytecodes::_fast_fgetfield:
1.2692 + __ movflt(xmm0, field);
1.2693 + break;
1.2694 + case Bytecodes::_fast_dgetfield:
1.2695 + __ movdbl(xmm0, field);
1.2696 + break;
1.2697 + default:
1.2698 + ShouldNotReachHere();
1.2699 + }
1.2700 + // [jk] not needed currently
1.2701 + // if (os::is_MP()) {
1.2702 + // Label notVolatile;
1.2703 + // __ testl(rdx, rdx);
1.2704 + // __ jcc(Assembler::zero, notVolatile);
1.2705 + // __ membar(Assembler::LoadLoad);
1.2706 + // __ bind(notVolatile);
1.2707 + //};
1.2708 +}
1.2709 +
1.2710 +void TemplateTable::fast_xaccess(TosState state) {
1.2711 + transition(vtos, state);
1.2712 +
1.2713 + // get receiver
1.2714 + __ movq(rax, aaddress(0));
1.2715 + debug_only(__ verify_local_tag(frame::TagReference, 0));
1.2716 + // access constant pool cache
1.2717 + __ get_cache_and_index_at_bcp(rcx, rdx, 2);
1.2718 + __ movq(rbx,
1.2719 + Address(rcx, rdx, Address::times_8,
1.2720 + in_bytes(constantPoolCacheOopDesc::base_offset() +
1.2721 + ConstantPoolCacheEntry::f2_offset())));
1.2722 + // make sure exception is reported in correct bcp range (getfield is
1.2723 + // next instruction)
1.2724 + __ incrementq(r13);
1.2725 + __ null_check(rax);
1.2726 + switch (state) {
1.2727 + case itos:
1.2728 + __ movl(rax, Address(rax, rbx, Address::times_1));
1.2729 + break;
1.2730 + case atos:
1.2731 + __ movq(rax, Address(rax, rbx, Address::times_1));
1.2732 + __ verify_oop(rax);
1.2733 + break;
1.2734 + case ftos:
1.2735 + __ movflt(xmm0, Address(rax, rbx, Address::times_1));
1.2736 + break;
1.2737 + default:
1.2738 + ShouldNotReachHere();
1.2739 + }
1.2740 +
1.2741 + // [jk] not needed currently
1.2742 + // if (os::is_MP()) {
1.2743 + // Label notVolatile;
1.2744 + // __ movl(rdx, Address(rcx, rdx, Address::times_8,
1.2745 + // in_bytes(constantPoolCacheOopDesc::base_offset() +
1.2746 + // ConstantPoolCacheEntry::flags_offset())));
1.2747 + // __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
1.2748 + // __ testl(rdx, 0x1);
1.2749 + // __ jcc(Assembler::zero, notVolatile);
1.2750 + // __ membar(Assembler::LoadLoad);
1.2751 + // __ bind(notVolatile);
1.2752 + // }
1.2753 +
1.2754 + __ decrementq(r13);
1.2755 +}
1.2756 +
1.2757 +
1.2758 +
1.2759 +//-----------------------------------------------------------------------------
1.2760 +// Calls
1.2761 +
1.2762 +void TemplateTable::count_calls(Register method, Register temp) {
1.2763 + // implemented elsewhere
1.2764 + ShouldNotReachHere();
1.2765 +}
1.2766 +
1.2767 +void TemplateTable::prepare_invoke(Register method,
1.2768 + Register index,
1.2769 + int byte_no,
1.2770 + Bytecodes::Code code) {
1.2771 + // determine flags
1.2772 + const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
1.2773 + const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
1.2774 + const bool is_invokespecial = code == Bytecodes::_invokespecial;
1.2775 + const bool load_receiver = code != Bytecodes::_invokestatic;
1.2776 + const bool receiver_null_check = is_invokespecial;
1.2777 + const bool save_flags = is_invokeinterface || is_invokevirtual;
1.2778 + // setup registers & access constant pool cache
1.2779 + const Register recv = rcx;
1.2780 + const Register flags = rdx;
1.2781 + assert_different_registers(method, index, recv, flags);
1.2782 +
1.2783 + // save 'interpreter return address'
1.2784 + __ save_bcp();
1.2785 +
1.2786 + load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual);
1.2787 +
1.2788 + // load receiver if needed (note: no return address pushed yet)
1.2789 + if (load_receiver) {
1.2790 + __ movl(recv, flags);
1.2791 + __ andl(recv, 0xFF);
1.2792 + if (TaggedStackInterpreter) __ shll(recv, 1); // index*2
1.2793 + __ movq(recv, Address(rsp, recv, Address::times_8, -Interpreter::expr_offset_in_bytes(1)));
1.2794 + __ verify_oop(recv);
1.2795 + }
1.2796 +
1.2797 + // do null check if needed
1.2798 + if (receiver_null_check) {
1.2799 + __ null_check(recv);
1.2800 + }
1.2801 +
1.2802 + if (save_flags) {
1.2803 + __ movl(r13, flags);
1.2804 + }
1.2805 +
1.2806 + // compute return type
1.2807 + __ shrl(flags, ConstantPoolCacheEntry::tosBits);
1.2808 + // Make sure we don't need to mask flags for tosBits after the above shift
1.2809 + ConstantPoolCacheEntry::verify_tosBits();
1.2810 + // load return address
1.2811 + {
1.2812 + ExternalAddress return_5((address)Interpreter::return_5_addrs_by_index_table());
1.2813 + ExternalAddress return_3((address)Interpreter::return_3_addrs_by_index_table());
1.2814 + __ lea(rscratch1, (is_invokeinterface ? return_5 : return_3));
1.2815 + __ movq(flags, Address(rscratch1, flags, Address::times_8));
1.2816 + }
1.2817 +
1.2818 + // push return address
1.2819 + __ pushq(flags);
1.2820 +
1.2821 + // Restore flag field from the constant pool cache, and restore esi
1.2822 + // for later null checks. r13 is the bytecode pointer
1.2823 + if (save_flags) {
1.2824 + __ movl(flags, r13);
1.2825 + __ restore_bcp();
1.2826 + }
1.2827 +}
1.2828 +
1.2829 +
1.2830 +void TemplateTable::invokevirtual_helper(Register index,
1.2831 + Register recv,
1.2832 + Register flags) {
1.2833 + // Uses temporary registers rax, rdx assert_different_registers(index, recv, rax, rdx);
1.2834 +
1.2835 + // Test for an invoke of a final method
1.2836 + Label notFinal;
1.2837 + __ movl(rax, flags);
1.2838 + __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod));
1.2839 + __ jcc(Assembler::zero, notFinal);
1.2840 +
1.2841 + const Register method = index; // method must be rbx
1.2842 + assert(method == rbx,
1.2843 + "methodOop must be rbx for interpreter calling convention");
1.2844 +
1.2845 + // do the call - the index is actually the method to call
1.2846 + __ verify_oop(method);
1.2847 +
1.2848 + // It's final, need a null check here!
1.2849 + __ null_check(recv);
1.2850 +
1.2851 + // profile this call
1.2852 + __ profile_final_call(rax);
1.2853 +
1.2854 + __ jump_from_interpreted(method, rax);
1.2855 +
1.2856 + __ bind(notFinal);
1.2857 +
1.2858 + // get receiver klass
1.2859 + __ null_check(recv, oopDesc::klass_offset_in_bytes());
1.2860 + __ movq(rax, Address(recv, oopDesc::klass_offset_in_bytes()));
1.2861 +
1.2862 + __ verify_oop(rax);
1.2863 +
1.2864 + // profile this call
1.2865 + __ profile_virtual_call(rax, r14, rdx);
1.2866 +
1.2867 + // get target methodOop & entry point
1.2868 + const int base = instanceKlass::vtable_start_offset() * wordSize;
1.2869 + assert(vtableEntry::size() * wordSize == 8,
1.2870 + "adjust the scaling in the code below");
1.2871 + __ movq(method, Address(rax, index,
1.2872 + Address::times_8,
1.2873 + base + vtableEntry::method_offset_in_bytes()));
1.2874 + __ movq(rdx, Address(method, methodOopDesc::interpreter_entry_offset()));
1.2875 + __ jump_from_interpreted(method, rdx);
1.2876 +}
1.2877 +
1.2878 +
1.2879 +void TemplateTable::invokevirtual(int byte_no) {
1.2880 + transition(vtos, vtos);
1.2881 + prepare_invoke(rbx, noreg, byte_no, bytecode());
1.2882 +
1.2883 + // rbx: index
1.2884 + // rcx: receiver
1.2885 + // rdx: flags
1.2886 +
1.2887 + invokevirtual_helper(rbx, rcx, rdx);
1.2888 +}
1.2889 +
1.2890 +
1.2891 +void TemplateTable::invokespecial(int byte_no) {
1.2892 + transition(vtos, vtos);
1.2893 + prepare_invoke(rbx, noreg, byte_no, bytecode());
1.2894 + // do the call
1.2895 + __ verify_oop(rbx);
1.2896 + __ profile_call(rax);
1.2897 + __ jump_from_interpreted(rbx, rax);
1.2898 +}
1.2899 +
1.2900 +
1.2901 +void TemplateTable::invokestatic(int byte_no) {
1.2902 + transition(vtos, vtos);
1.2903 + prepare_invoke(rbx, noreg, byte_no, bytecode());
1.2904 + // do the call
1.2905 + __ verify_oop(rbx);
1.2906 + __ profile_call(rax);
1.2907 + __ jump_from_interpreted(rbx, rax);
1.2908 +}
1.2909 +
1.2910 +void TemplateTable::fast_invokevfinal(int byte_no) {
1.2911 + transition(vtos, vtos);
1.2912 + __ stop("fast_invokevfinal not used on amd64");
1.2913 +}
1.2914 +
1.2915 +void TemplateTable::invokeinterface(int byte_no) {
1.2916 + transition(vtos, vtos);
1.2917 + prepare_invoke(rax, rbx, byte_no, bytecode());
1.2918 +
1.2919 + // rax: Interface
1.2920 + // rbx: index
1.2921 + // rcx: receiver
1.2922 + // rdx: flags
1.2923 +
1.2924 + // Special case of invokeinterface called for virtual method of
1.2925 + // java.lang.Object. See cpCacheOop.cpp for details.
1.2926 + // This code isn't produced by javac, but could be produced by
1.2927 + // another compliant java compiler.
1.2928 + Label notMethod;
1.2929 + __ movl(r14, rdx);
1.2930 + __ andl(r14, (1 << ConstantPoolCacheEntry::methodInterface));
1.2931 + __ jcc(Assembler::zero, notMethod);
1.2932 +
1.2933 + invokevirtual_helper(rbx, rcx, rdx);
1.2934 + __ bind(notMethod);
1.2935 +
1.2936 + // Get receiver klass into rdx - also a null check
1.2937 + __ restore_locals(); // restore r14
1.2938 + __ movq(rdx, Address(rcx, oopDesc::klass_offset_in_bytes()));
1.2939 + __ verify_oop(rdx);
1.2940 +
1.2941 + // profile this call
1.2942 + __ profile_virtual_call(rdx, r13, r14);
1.2943 +
1.2944 + __ movq(r14, rdx); // Save klassOop in r14
1.2945 +
1.2946 + // Compute start of first itableOffsetEntry (which is at the end of
1.2947 + // the vtable)
1.2948 + const int base = instanceKlass::vtable_start_offset() * wordSize;
1.2949 + // Get length of vtable
1.2950 + assert(vtableEntry::size() * wordSize == 8,
1.2951 + "adjust the scaling in the code below");
1.2952 + __ movl(r13, Address(rdx,
1.2953 + instanceKlass::vtable_length_offset() * wordSize));
1.2954 + __ leaq(rdx, Address(rdx, r13, Address::times_8, base));
1.2955 +
1.2956 + if (HeapWordsPerLong > 1) {
1.2957 + // Round up to align_object_offset boundary
1.2958 + __ round_to_q(rdx, BytesPerLong);
1.2959 + }
1.2960 +
1.2961 + Label entry, search, interface_ok;
1.2962 +
1.2963 + __ jmpb(entry);
1.2964 + __ bind(search);
1.2965 + __ addq(rdx, itableOffsetEntry::size() * wordSize);
1.2966 +
1.2967 + __ bind(entry);
1.2968 +
1.2969 + // Check that the entry is non-null. A null entry means that the
1.2970 + // receiver class doesn't implement the interface, and wasn't the
1.2971 + // same as the receiver class checked when the interface was
1.2972 + // resolved.
1.2973 + __ pushq(rdx);
1.2974 + __ movq(rdx, Address(rdx, itableOffsetEntry::interface_offset_in_bytes()));
1.2975 + __ testq(rdx, rdx);
1.2976 + __ jcc(Assembler::notZero, interface_ok);
1.2977 + // throw exception
1.2978 + __ popq(rdx); // pop saved register first.
1.2979 + __ popq(rbx); // pop return address (pushed by prepare_invoke)
1.2980 + __ restore_bcp(); // r13 must be correct for exception handler (was
1.2981 + // destroyed)
1.2982 + __ restore_locals(); // make sure locals pointer is correct as well
1.2983 + // (was destroyed)
1.2984 + __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1.2985 + InterpreterRuntime::throw_IncompatibleClassChangeError));
1.2986 + // the call_VM checks for exception, so we should never return here.
1.2987 + __ should_not_reach_here();
1.2988 + __ bind(interface_ok);
1.2989 +
1.2990 + __ popq(rdx);
1.2991 +
1.2992 + __ cmpq(rax, Address(rdx, itableOffsetEntry::interface_offset_in_bytes()));
1.2993 + __ jcc(Assembler::notEqual, search);
1.2994 +
1.2995 + __ movl(rdx, Address(rdx, itableOffsetEntry::offset_offset_in_bytes()));
1.2996 +
1.2997 + __ addq(rdx, r14); // Add offset to klassOop
1.2998 + assert(itableMethodEntry::size() * wordSize == 8,
1.2999 + "adjust the scaling in the code below");
1.3000 + __ movq(rbx, Address(rdx, rbx, Address::times_8));
1.3001 + // rbx: methodOop to call
1.3002 + // rcx: receiver
1.3003 + // Check for abstract method error
1.3004 + // Note: This should be done more efficiently via a
1.3005 + // throw_abstract_method_error interpreter entry point and a
1.3006 + // conditional jump to it in case of a null method.
1.3007 + {
1.3008 + Label L;
1.3009 + __ testq(rbx, rbx);
1.3010 + __ jcc(Assembler::notZero, L);
1.3011 + // throw exception
1.3012 + // note: must restore interpreter registers to canonical
1.3013 + // state for exception handling to work correctly!
1.3014 + __ popq(rbx); // pop return address (pushed by prepare_invoke)
1.3015 + __ restore_bcp(); // r13 must be correct for exception handler
1.3016 + // (was destroyed)
1.3017 + __ restore_locals(); // make sure locals pointer is correct as
1.3018 + // well (was destroyed)
1.3019 + __ call_VM(noreg,
1.3020 + CAST_FROM_FN_PTR(address,
1.3021 + InterpreterRuntime::throw_AbstractMethodError));
1.3022 + // the call_VM checks for exception, so we should never return here.
1.3023 + __ should_not_reach_here();
1.3024 + __ bind(L);
1.3025 + }
1.3026 +
1.3027 + __ movq(rcx, Address(rbx, methodOopDesc::interpreter_entry_offset()));
1.3028 +
1.3029 + // do the call
1.3030 + // rcx: receiver
1.3031 + // rbx: methodOop
1.3032 + __ jump_from_interpreted(rbx, rdx);
1.3033 +}
1.3034 +
1.3035 +//-----------------------------------------------------------------------------
1.3036 +// Allocation
1.3037 +
1.3038 +void TemplateTable::_new() {
1.3039 + transition(vtos, atos);
1.3040 + __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
1.3041 + Label slow_case;
1.3042 + Label done;
1.3043 + Label initialize_header;
1.3044 + Label initialize_object; // including clearing the fields
1.3045 + Label allocate_shared;
1.3046 + ExternalAddress top((address)Universe::heap()->top_addr());
1.3047 + ExternalAddress end((address)Universe::heap()->end_addr());
1.3048 +
1.3049 + __ get_cpool_and_tags(rsi, rax);
1.3050 + // get instanceKlass
1.3051 + __ movq(rsi, Address(rsi, rdx,
1.3052 + Address::times_8, sizeof(constantPoolOopDesc)));
1.3053 +
1.3054 + // make sure the class we're about to instantiate has been
1.3055 + // resolved. Note: slow_case does a pop of stack, which is why we
1.3056 + // loaded class/pushed above
1.3057 + const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
1.3058 + __ cmpb(Address(rax, rdx, Address::times_1, tags_offset),
1.3059 + JVM_CONSTANT_Class);
1.3060 + __ jcc(Assembler::notEqual, slow_case);
1.3061 +
1.3062 + // make sure klass is initialized & doesn't have finalizer
1.3063 + // make sure klass is fully initialized
1.3064 + __ cmpl(Address(rsi,
1.3065 + instanceKlass::init_state_offset_in_bytes() +
1.3066 + sizeof(oopDesc)),
1.3067 + instanceKlass::fully_initialized);
1.3068 + __ jcc(Assembler::notEqual, slow_case);
1.3069 +
1.3070 + // get instance_size in instanceKlass (scaled to a count of bytes)
1.3071 + __ movl(rdx,
1.3072 + Address(rsi,
1.3073 + Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
1.3074 + // test to see if it has a finalizer or is malformed in some way
1.3075 + __ testl(rdx, Klass::_lh_instance_slow_path_bit);
1.3076 + __ jcc(Assembler::notZero, slow_case);
1.3077 +
1.3078 + // Allocate the instance
1.3079 + // 1) Try to allocate in the TLAB
1.3080 + // 2) if fail and the object is large allocate in the shared Eden
1.3081 + // 3) if the above fails (or is not applicable), go to a slow case
1.3082 + // (creates a new TLAB, etc.)
1.3083 +
1.3084 + const bool allow_shared_alloc =
1.3085 + Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
1.3086 +
1.3087 + if (UseTLAB) {
1.3088 + __ movq(rax, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
1.3089 + __ leaq(rbx, Address(rax, rdx, Address::times_1));
1.3090 + __ cmpq(rbx, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
1.3091 + __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
1.3092 + __ movq(Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
1.3093 + if (ZeroTLAB) {
1.3094 + // the fields have been already cleared
1.3095 + __ jmp(initialize_header);
1.3096 + } else {
1.3097 + // initialize both the header and fields
1.3098 + __ jmp(initialize_object);
1.3099 + }
1.3100 + }
1.3101 +
1.3102 + // Allocation in the shared Eden, if allowed.
1.3103 + //
1.3104 + // rdx: instance size in bytes
1.3105 + if (allow_shared_alloc) {
1.3106 + __ bind(allocate_shared);
1.3107 +
1.3108 + const Register RtopAddr = rscratch1;
1.3109 + const Register RendAddr = rscratch2;
1.3110 +
1.3111 + __ lea(RtopAddr, top);
1.3112 + __ lea(RendAddr, end);
1.3113 + __ movq(rax, Address(RtopAddr, 0));
1.3114 +
1.3115 + // For retries rax gets set by cmpxchgq
1.3116 + Label retry;
1.3117 + __ bind(retry);
1.3118 + __ leaq(rbx, Address(rax, rdx, Address::times_1));
1.3119 + __ cmpq(rbx, Address(RendAddr, 0));
1.3120 + __ jcc(Assembler::above, slow_case);
1.3121 +
1.3122 + // Compare rax with the top addr, and if still equal, store the new
1.3123 + // top addr in rbx at the address of the top addr pointer. Sets ZF if was
1.3124 + // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
1.3125 + //
1.3126 + // rax: object begin
1.3127 + // rbx: object end
1.3128 + // rdx: instance size in bytes
1.3129 + if (os::is_MP()) {
1.3130 + __ lock();
1.3131 + }
1.3132 + __ cmpxchgq(rbx, Address(RtopAddr, 0));
1.3133 +
1.3134 + // if someone beat us on the allocation, try again, otherwise continue
1.3135 + __ jcc(Assembler::notEqual, retry);
1.3136 + }
1.3137 +
1.3138 + if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
1.3139 + // The object is initialized before the header. If the object size is
1.3140 + // zero, go directly to the header initialization.
1.3141 + __ bind(initialize_object);
1.3142 + __ decrementl(rdx, sizeof(oopDesc));
1.3143 + __ jcc(Assembler::zero, initialize_header);
1.3144 +
1.3145 + // Initialize object fields
1.3146 + __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
1.3147 + __ shrl(rdx, LogBytesPerLong); // divide by oopSize to simplify the loop
1.3148 + {
1.3149 + Label loop;
1.3150 + __ bind(loop);
1.3151 + __ movq(Address(rax, rdx, Address::times_8,
1.3152 + sizeof(oopDesc) - oopSize),
1.3153 + rcx);
1.3154 + __ decrementl(rdx);
1.3155 + __ jcc(Assembler::notZero, loop);
1.3156 + }
1.3157 +
1.3158 + // initialize object header only.
1.3159 + __ bind(initialize_header);
1.3160 + if (UseBiasedLocking) {
1.3161 + __ movq(rscratch1, Address(rsi, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
1.3162 + __ movq(Address(rax, oopDesc::mark_offset_in_bytes()), rscratch1);
1.3163 + } else {
1.3164 + __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()),
1.3165 + (intptr_t) markOopDesc::prototype()); // header (address 0x1)
1.3166 + }
1.3167 + __ movq(Address(rax, oopDesc::klass_offset_in_bytes()), rsi); // klass
1.3168 + __ jmp(done);
1.3169 + }
1.3170 +
1.3171 + {
1.3172 + SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
1.3173 + // Trigger dtrace event for fastpath
1.3174 + __ push(atos); // save the return value
1.3175 + __ call_VM_leaf(
1.3176 + CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
1.3177 + __ pop(atos); // restore the return value
1.3178 + }
1.3179 +
1.3180 + // slow case
1.3181 + __ bind(slow_case);
1.3182 + __ get_constant_pool(c_rarg1);
1.3183 + __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
1.3184 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
1.3185 + __ verify_oop(rax);
1.3186 +
1.3187 + // continue
1.3188 + __ bind(done);
1.3189 +}
1.3190 +
1.3191 +void TemplateTable::newarray() {
1.3192 + transition(itos, atos);
1.3193 + __ load_unsigned_byte(c_rarg1, at_bcp(1));
1.3194 + __ movl(c_rarg2, rax);
1.3195 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
1.3196 + c_rarg1, c_rarg2);
1.3197 +}
1.3198 +
1.3199 +void TemplateTable::anewarray() {
1.3200 + transition(itos, atos);
1.3201 + __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
1.3202 + __ get_constant_pool(c_rarg1);
1.3203 + __ movl(c_rarg3, rax);
1.3204 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
1.3205 + c_rarg1, c_rarg2, c_rarg3);
1.3206 +}
1.3207 +
1.3208 +void TemplateTable::arraylength() {
1.3209 + transition(atos, itos);
1.3210 + __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
1.3211 + __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
1.3212 +}
1.3213 +
1.3214 +void TemplateTable::checkcast() {
1.3215 + transition(atos, atos);
1.3216 + Label done, is_null, ok_is_subtype, quicked, resolved;
1.3217 + __ testq(rax, rax); // object is in rax
1.3218 + __ jcc(Assembler::zero, is_null);
1.3219 +
1.3220 + // Get cpool & tags index
1.3221 + __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
1.3222 + __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
1.3223 + // See if bytecode has already been quicked
1.3224 + __ cmpb(Address(rdx, rbx,
1.3225 + Address::times_1,
1.3226 + typeArrayOopDesc::header_size(T_BYTE) * wordSize),
1.3227 + JVM_CONSTANT_Class);
1.3228 + __ jcc(Assembler::equal, quicked);
1.3229 +
1.3230 + __ movq(r12, rcx); // save rcx XXX
1.3231 + __ push(atos); // save receiver for result, and for GC
1.3232 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
1.3233 + __ pop_ptr(rdx); // restore receiver
1.3234 + __ movq(rcx, r12); // restore rcx XXX
1.3235 + __ jmpb(resolved);
1.3236 +
1.3237 + // Get superklass in rax and subklass in rbx
1.3238 + __ bind(quicked);
1.3239 + __ movq(rdx, rax); // Save object in rdx; rax needed for subtype check
1.3240 + __ movq(rax, Address(rcx, rbx,
1.3241 + Address::times_8, sizeof(constantPoolOopDesc)));
1.3242 +
1.3243 + __ bind(resolved);
1.3244 + __ movq(rbx, Address(rdx, oopDesc::klass_offset_in_bytes()));
1.3245 +
1.3246 + // Generate subtype check. Blows rcx, rdi. Object in rdx.
1.3247 + // Superklass in rax. Subklass in rbx.
1.3248 + __ gen_subtype_check(rbx, ok_is_subtype);
1.3249 +
1.3250 + // Come here on failure
1.3251 + __ push_ptr(rdx);
1.3252 + // object is at TOS
1.3253 + __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
1.3254 +
1.3255 + // Come here on success
1.3256 + __ bind(ok_is_subtype);
1.3257 + __ movq(rax, rdx); // Restore object in rdx
1.3258 +
1.3259 + // Collect counts on whether this check-cast sees NULLs a lot or not.
1.3260 + if (ProfileInterpreter) {
1.3261 + __ jmp(done);
1.3262 + __ bind(is_null);
1.3263 + __ profile_null_seen(rcx);
1.3264 + } else {
1.3265 + __ bind(is_null); // same as 'done'
1.3266 + }
1.3267 + __ bind(done);
1.3268 +}
1.3269 +
1.3270 +void TemplateTable::instanceof() {
1.3271 + transition(atos, itos);
1.3272 + Label done, is_null, ok_is_subtype, quicked, resolved;
1.3273 + __ testq(rax, rax);
1.3274 + __ jcc(Assembler::zero, is_null);
1.3275 +
1.3276 + // Get cpool & tags index
1.3277 + __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
1.3278 + __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
1.3279 + // See if bytecode has already been quicked
1.3280 + __ cmpb(Address(rdx, rbx,
1.3281 + Address::times_1,
1.3282 + typeArrayOopDesc::header_size(T_BYTE) * wordSize),
1.3283 + JVM_CONSTANT_Class);
1.3284 + __ jcc(Assembler::equal, quicked);
1.3285 +
1.3286 + __ movq(r12, rcx); // save rcx
1.3287 + __ push(atos); // save receiver for result, and for GC
1.3288 + call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
1.3289 + __ pop_ptr(rdx); // restore receiver
1.3290 + __ movq(rdx, Address(rdx, oopDesc::klass_offset_in_bytes()));
1.3291 + __ movq(rcx, r12); // restore rcx
1.3292 + __ jmpb(resolved);
1.3293 +
1.3294 + // Get superklass in rax and subklass in rdx
1.3295 + __ bind(quicked);
1.3296 + __ movq(rdx, Address(rax, oopDesc::klass_offset_in_bytes()));
1.3297 + __ movq(rax, Address(rcx, rbx,
1.3298 + Address::times_8, sizeof(constantPoolOopDesc)));
1.3299 +
1.3300 + __ bind(resolved);
1.3301 +
1.3302 + // Generate subtype check. Blows rcx, rdi
1.3303 + // Superklass in rax. Subklass in rdx.
1.3304 + __ gen_subtype_check(rdx, ok_is_subtype);
1.3305 +
1.3306 + // Come here on failure
1.3307 + __ xorl(rax, rax);
1.3308 + __ jmpb(done);
1.3309 + // Come here on success
1.3310 + __ bind(ok_is_subtype);
1.3311 + __ movl(rax, 1);
1.3312 +
1.3313 + // Collect counts on whether this test sees NULLs a lot or not.
1.3314 + if (ProfileInterpreter) {
1.3315 + __ jmp(done);
1.3316 + __ bind(is_null);
1.3317 + __ profile_null_seen(rcx);
1.3318 + } else {
1.3319 + __ bind(is_null); // same as 'done'
1.3320 + }
1.3321 + __ bind(done);
1.3322 + // rax = 0: obj == NULL or obj is not an instanceof the specified klass
1.3323 + // rax = 1: obj != NULL and obj is an instanceof the specified klass
1.3324 +}
1.3325 +
1.3326 +//-----------------------------------------------------------------------------
1.3327 +// Breakpoints
1.3328 +void TemplateTable::_breakpoint() {
1.3329 + // Note: We get here even if we are single stepping..
1.3330 + // jbug inists on setting breakpoints at every bytecode
1.3331 + // even if we are in single step mode.
1.3332 +
1.3333 + transition(vtos, vtos);
1.3334 +
1.3335 + // get the unpatched byte code
1.3336 + __ get_method(c_rarg1);
1.3337 + __ call_VM(noreg,
1.3338 + CAST_FROM_FN_PTR(address,
1.3339 + InterpreterRuntime::get_original_bytecode_at),
1.3340 + c_rarg1, r13);
1.3341 + __ movq(rbx, rax);
1.3342 +
1.3343 + // post the breakpoint event
1.3344 + __ get_method(c_rarg1);
1.3345 + __ call_VM(noreg,
1.3346 + CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
1.3347 + c_rarg1, r13);
1.3348 +
1.3349 + // complete the execution of original bytecode
1.3350 + __ dispatch_only_normal(vtos);
1.3351 +}
1.3352 +
1.3353 +//-----------------------------------------------------------------------------
1.3354 +// Exceptions
1.3355 +
1.3356 +void TemplateTable::athrow() {
1.3357 + transition(atos, vtos);
1.3358 + __ null_check(rax);
1.3359 + __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
1.3360 +}
1.3361 +
1.3362 +//-----------------------------------------------------------------------------
1.3363 +// Synchronization
1.3364 +//
1.3365 +// Note: monitorenter & exit are symmetric routines; which is reflected
1.3366 +// in the assembly code structure as well
1.3367 +//
1.3368 +// Stack layout:
1.3369 +//
1.3370 +// [expressions ] <--- rsp = expression stack top
1.3371 +// ..
1.3372 +// [expressions ]
1.3373 +// [monitor entry] <--- monitor block top = expression stack bot
1.3374 +// ..
1.3375 +// [monitor entry]
1.3376 +// [frame data ] <--- monitor block bot
1.3377 +// ...
1.3378 +// [saved rbp ] <--- rbp
1.3379 +void TemplateTable::monitorenter() {
1.3380 + transition(atos, vtos);
1.3381 +
1.3382 + // check for NULL object
1.3383 + __ null_check(rax);
1.3384 +
1.3385 + const Address monitor_block_top(
1.3386 + rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1.3387 + const Address monitor_block_bot(
1.3388 + rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1.3389 + const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1.3390 +
1.3391 + Label allocated;
1.3392 +
1.3393 + // initialize entry pointer
1.3394 + __ xorl(c_rarg1, c_rarg1); // points to free slot or NULL
1.3395 +
1.3396 + // find a free slot in the monitor block (result in c_rarg1)
1.3397 + {
1.3398 + Label entry, loop, exit;
1.3399 + __ movq(c_rarg3, monitor_block_top); // points to current entry,
1.3400 + // starting with top-most entry
1.3401 + __ leaq(c_rarg2, monitor_block_bot); // points to word before bottom
1.3402 + // of monitor block
1.3403 + __ jmpb(entry);
1.3404 +
1.3405 + __ bind(loop);
1.3406 + // check if current entry is used
1.3407 + __ cmpq(Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()), (int) NULL);
1.3408 + // if not used then remember entry in c_rarg1
1.3409 + __ cmovq(Assembler::equal, c_rarg1, c_rarg3);
1.3410 + // check if current entry is for same object
1.3411 + __ cmpq(rax, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
1.3412 + // if same object then stop searching
1.3413 + __ jccb(Assembler::equal, exit);
1.3414 + // otherwise advance to next entry
1.3415 + __ addq(c_rarg3, entry_size);
1.3416 + __ bind(entry);
1.3417 + // check if bottom reached
1.3418 + __ cmpq(c_rarg3, c_rarg2);
1.3419 + // if not at bottom then check this entry
1.3420 + __ jcc(Assembler::notEqual, loop);
1.3421 + __ bind(exit);
1.3422 + }
1.3423 +
1.3424 + __ testq(c_rarg1, c_rarg1); // check if a slot has been found
1.3425 + __ jcc(Assembler::notZero, allocated); // if found, continue with that one
1.3426 +
1.3427 + // allocate one if there's no free slot
1.3428 + {
1.3429 + Label entry, loop;
1.3430 + // 1. compute new pointers // rsp: old expression stack top
1.3431 + __ movq(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
1.3432 + __ subq(rsp, entry_size); // move expression stack top
1.3433 + __ subq(c_rarg1, entry_size); // move expression stack bottom
1.3434 + __ movq(c_rarg3, rsp); // set start value for copy loop
1.3435 + __ movq(monitor_block_bot, c_rarg1); // set new monitor block bottom
1.3436 + __ jmp(entry);
1.3437 + // 2. move expression stack contents
1.3438 + __ bind(loop);
1.3439 + __ movq(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
1.3440 + // word from old location
1.3441 + __ movq(Address(c_rarg3, 0), c_rarg2); // and store it at new location
1.3442 + __ addq(c_rarg3, wordSize); // advance to next word
1.3443 + __ bind(entry);
1.3444 + __ cmpq(c_rarg3, c_rarg1); // check if bottom reached
1.3445 + __ jcc(Assembler::notEqual, loop); // if not at bottom then
1.3446 + // copy next word
1.3447 + }
1.3448 +
1.3449 + // call run-time routine
1.3450 + // c_rarg1: points to monitor entry
1.3451 + __ bind(allocated);
1.3452 +
1.3453 + // Increment bcp to point to the next bytecode, so exception
1.3454 + // handling for async. exceptions work correctly.
1.3455 + // The object has already been poped from the stack, so the
1.3456 + // expression stack looks correct.
1.3457 + __ incrementq(r13);
1.3458 +
1.3459 + // store object
1.3460 + __ movq(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), rax);
1.3461 + __ lock_object(c_rarg1);
1.3462 +
1.3463 + // check to make sure this monitor doesn't cause stack overflow after locking
1.3464 + __ save_bcp(); // in case of exception
1.3465 + __ generate_stack_overflow_check(0);
1.3466 +
1.3467 + // The bcp has already been incremented. Just need to dispatch to
1.3468 + // next instruction.
1.3469 + __ dispatch_next(vtos);
1.3470 +}
1.3471 +
1.3472 +
1.3473 +void TemplateTable::monitorexit() {
1.3474 + transition(atos, vtos);
1.3475 +
1.3476 + // check for NULL object
1.3477 + __ null_check(rax);
1.3478 +
1.3479 + const Address monitor_block_top(
1.3480 + rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1.3481 + const Address monitor_block_bot(
1.3482 + rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1.3483 + const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1.3484 +
1.3485 + Label found;
1.3486 +
1.3487 + // find matching slot
1.3488 + {
1.3489 + Label entry, loop;
1.3490 + __ movq(c_rarg1, monitor_block_top); // points to current entry,
1.3491 + // starting with top-most entry
1.3492 + __ leaq(c_rarg2, monitor_block_bot); // points to word before bottom
1.3493 + // of monitor block
1.3494 + __ jmpb(entry);
1.3495 +
1.3496 + __ bind(loop);
1.3497 + // check if current entry is for same object
1.3498 + __ cmpq(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1.3499 + // if same object then stop searching
1.3500 + __ jcc(Assembler::equal, found);
1.3501 + // otherwise advance to next entry
1.3502 + __ addq(c_rarg1, entry_size);
1.3503 + __ bind(entry);
1.3504 + // check if bottom reached
1.3505 + __ cmpq(c_rarg1, c_rarg2);
1.3506 + // if not at bottom then check this entry
1.3507 + __ jcc(Assembler::notEqual, loop);
1.3508 + }
1.3509 +
1.3510 + // error handling. Unlocking was not block-structured
1.3511 + __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1.3512 + InterpreterRuntime::throw_illegal_monitor_state_exception));
1.3513 + __ should_not_reach_here();
1.3514 +
1.3515 + // call run-time routine
1.3516 + // rsi: points to monitor entry
1.3517 + __ bind(found);
1.3518 + __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
1.3519 + __ unlock_object(c_rarg1);
1.3520 + __ pop_ptr(rax); // discard object
1.3521 +}
1.3522 +
1.3523 +
1.3524 +// Wide instructions
1.3525 +void TemplateTable::wide() {
1.3526 + transition(vtos, vtos);
1.3527 + __ load_unsigned_byte(rbx, at_bcp(1));
1.3528 + __ lea(rscratch1, ExternalAddress((address)Interpreter::_wentry_point));
1.3529 + __ jmp(Address(rscratch1, rbx, Address::times_8));
1.3530 + // Note: the r13 increment step is part of the individual wide
1.3531 + // bytecode implementations
1.3532 +}
1.3533 +
1.3534 +
1.3535 +// Multi arrays
1.3536 +void TemplateTable::multianewarray() {
1.3537 + transition(vtos, atos);
1.3538 + __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
1.3539 + // last dim is on top of stack; we want address of first one:
1.3540 + // first_addr = last_addr + (ndims - 1) * wordSize
1.3541 + if (TaggedStackInterpreter) __ shll(rax, 1); // index*2
1.3542 + __ leaq(c_rarg1, Address(rsp, rax, Address::times_8, -wordSize));
1.3543 + call_VM(rax,
1.3544 + CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
1.3545 + c_rarg1);
1.3546 + __ load_unsigned_byte(rbx, at_bcp(3));
1.3547 + if (TaggedStackInterpreter) __ shll(rbx, 1); // index*2
1.3548 + __ leaq(rsp, Address(rsp, rbx, Address::times_8));
1.3549 +}
