src/cpu/x86/vm/jniFastGetField_x86_32.cpp

changeset 0
f90c822e73f8
child 6876
710a3c8b516e
     1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/src/cpu/x86/vm/jniFastGetField_x86_32.cpp	Wed Apr 27 01:25:04 2016 +0800
     1.3 @@ -0,0 +1,368 @@
     1.4 +/*
     1.5 + * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
     1.6 + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
     1.7 + *
     1.8 + * This code is free software; you can redistribute it and/or modify it
     1.9 + * under the terms of the GNU General Public License version 2 only, as
    1.10 + * published by the Free Software Foundation.
    1.11 + *
    1.12 + * This code is distributed in the hope that it will be useful, but WITHOUT
    1.13 + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
    1.14 + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
    1.15 + * version 2 for more details (a copy is included in the LICENSE file that
    1.16 + * accompanied this code).
    1.17 + *
    1.18 + * You should have received a copy of the GNU General Public License version
    1.19 + * 2 along with this work; if not, write to the Free Software Foundation,
    1.20 + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
    1.21 + *
    1.22 + * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
    1.23 + * or visit www.oracle.com if you need additional information or have any
    1.24 + * questions.
    1.25 + *
    1.26 + */
    1.27 +
    1.28 +#include "precompiled.hpp"
    1.29 +#include "asm/macroAssembler.hpp"
    1.30 +#include "memory/resourceArea.hpp"
    1.31 +#include "prims/jniFastGetField.hpp"
    1.32 +#include "prims/jvm_misc.hpp"
    1.33 +#include "runtime/safepoint.hpp"
    1.34 +
    1.35 +#define __ masm->
    1.36 +
    1.37 +#define BUFFER_SIZE 30
    1.38 +
    1.39 +#ifdef _WINDOWS
    1.40 +GetBooleanField_t JNI_FastGetField::jni_fast_GetBooleanField_fp;
    1.41 +GetByteField_t    JNI_FastGetField::jni_fast_GetByteField_fp;
    1.42 +GetCharField_t    JNI_FastGetField::jni_fast_GetCharField_fp;
    1.43 +GetShortField_t   JNI_FastGetField::jni_fast_GetShortField_fp;
    1.44 +GetIntField_t     JNI_FastGetField::jni_fast_GetIntField_fp;
    1.45 +GetLongField_t    JNI_FastGetField::jni_fast_GetLongField_fp;
    1.46 +GetFloatField_t   JNI_FastGetField::jni_fast_GetFloatField_fp;
    1.47 +GetDoubleField_t  JNI_FastGetField::jni_fast_GetDoubleField_fp;
    1.48 +#endif
    1.49 +
    1.50 +// Instead of issuing lfence for LoadLoad barrier, we create data dependency
    1.51 +// between loads, which is much more efficient than lfence.
    1.52 +
    1.53 +address JNI_FastGetField::generate_fast_get_int_field0(BasicType type) {
    1.54 +  const char *name;
    1.55 +  switch (type) {
    1.56 +    case T_BOOLEAN: name = "jni_fast_GetBooleanField"; break;
    1.57 +    case T_BYTE:    name = "jni_fast_GetByteField";    break;
    1.58 +    case T_CHAR:    name = "jni_fast_GetCharField";    break;
    1.59 +    case T_SHORT:   name = "jni_fast_GetShortField";   break;
    1.60 +    case T_INT:     name = "jni_fast_GetIntField";     break;
    1.61 +    default:        ShouldNotReachHere();
    1.62 +  }
    1.63 +  ResourceMark rm;
    1.64 +  BufferBlob* blob = BufferBlob::create(name, BUFFER_SIZE*wordSize);
    1.65 +  CodeBuffer cbuf(blob);
    1.66 +  MacroAssembler* masm = new MacroAssembler(&cbuf);
    1.67 +  address fast_entry = __ pc();
    1.68 +
    1.69 +  Label slow;
    1.70 +
    1.71 +  // stack layout:    offset from rsp (in words):
    1.72 +  //  return pc        0
    1.73 +  //  jni env          1
    1.74 +  //  obj              2
    1.75 +  //  jfieldID         3
    1.76 +
    1.77 +  ExternalAddress counter(SafepointSynchronize::safepoint_counter_addr());
    1.78 +  __ mov32 (rcx, counter);
    1.79 +  __ testb (rcx, 1);
    1.80 +  __ jcc (Assembler::notZero, slow);
    1.81 +  if (os::is_MP()) {
    1.82 +    __ mov(rax, rcx);
    1.83 +    __ andptr(rax, 1);                         // rax, must end up 0
    1.84 +    __ movptr(rdx, Address(rsp, rax, Address::times_1, 2*wordSize));
    1.85 +                                              // obj, notice rax, is 0.
    1.86 +                                              // rdx is data dependent on rcx.
    1.87 +  } else {
    1.88 +    __ movptr (rdx, Address(rsp, 2*wordSize));  // obj
    1.89 +  }
    1.90 +  __ movptr(rax, Address(rsp, 3*wordSize));  // jfieldID
    1.91 +  __ movptr(rdx, Address(rdx, 0));           // *obj
    1.92 +  __ shrptr (rax, 2);                         // offset
    1.93 +
    1.94 +  assert(count < LIST_CAPACITY, "LIST_CAPACITY too small");
    1.95 +  speculative_load_pclist[count] = __ pc();
    1.96 +  switch (type) {
    1.97 +    case T_BOOLEAN: __ movzbl (rax, Address(rdx, rax, Address::times_1)); break;
    1.98 +    case T_BYTE:    __ movsbl (rax, Address(rdx, rax, Address::times_1)); break;
    1.99 +    case T_CHAR:    __ movzwl (rax, Address(rdx, rax, Address::times_1)); break;
   1.100 +    case T_SHORT:   __ movswl (rax, Address(rdx, rax, Address::times_1)); break;
   1.101 +    case T_INT:     __ movl   (rax, Address(rdx, rax, Address::times_1)); break;
   1.102 +    default:        ShouldNotReachHere();
   1.103 +  }
   1.104 +
   1.105 +  Address ca1;
   1.106 +  if (os::is_MP()) {
   1.107 +    __ lea(rdx, counter);
   1.108 +    __ xorptr(rdx, rax);
   1.109 +    __ xorptr(rdx, rax);
   1.110 +    __ cmp32(rcx, Address(rdx, 0));
   1.111 +    // ca1 is the same as ca because
   1.112 +    // rax, ^ counter_addr ^ rax, = address
   1.113 +    // ca1 is data dependent on rax,.
   1.114 +  } else {
   1.115 +    __ cmp32(rcx, counter);
   1.116 +  }
   1.117 +  __ jcc (Assembler::notEqual, slow);
   1.118 +
   1.119 +#ifndef _WINDOWS
   1.120 +  __ ret (0);
   1.121 +#else
   1.122 +  // __stdcall calling convention
   1.123 +  __ ret (3*wordSize);
   1.124 +#endif
   1.125 +
   1.126 +  slowcase_entry_pclist[count++] = __ pc();
   1.127 +  __ bind (slow);
   1.128 +  address slow_case_addr;
   1.129 +  switch (type) {
   1.130 +    case T_BOOLEAN: slow_case_addr = jni_GetBooleanField_addr(); break;
   1.131 +    case T_BYTE:    slow_case_addr = jni_GetByteField_addr();    break;
   1.132 +    case T_CHAR:    slow_case_addr = jni_GetCharField_addr();    break;
   1.133 +    case T_SHORT:   slow_case_addr = jni_GetShortField_addr();   break;
   1.134 +    case T_INT:     slow_case_addr = jni_GetIntField_addr();
   1.135 +  }
   1.136 +  // tail call
   1.137 +  __ jump (ExternalAddress(slow_case_addr));
   1.138 +
   1.139 +  __ flush ();
   1.140 +
   1.141 +#ifndef _WINDOWS
   1.142 +  return fast_entry;
   1.143 +#else
   1.144 +  switch (type) {
   1.145 +  case T_BOOLEAN: jni_fast_GetBooleanField_fp = (GetBooleanField_t) fast_entry; break;
   1.146 +  case T_BYTE:    jni_fast_GetByteField_fp    = (GetByteField_t)    fast_entry; break;
   1.147 +  case T_CHAR:    jni_fast_GetCharField_fp    = (GetCharField_t)    fast_entry; break;
   1.148 +  case T_SHORT:   jni_fast_GetShortField_fp   = (GetShortField_t)   fast_entry; break;
   1.149 +  case T_INT:     jni_fast_GetIntField_fp     = (GetIntField_t)     fast_entry; break;
   1.150 +  }
   1.151 +  return os::win32::fast_jni_accessor_wrapper(type);
   1.152 +#endif
   1.153 +}
   1.154 +
   1.155 +address JNI_FastGetField::generate_fast_get_boolean_field() {
   1.156 +  return generate_fast_get_int_field0(T_BOOLEAN);
   1.157 +}
   1.158 +
   1.159 +address JNI_FastGetField::generate_fast_get_byte_field() {
   1.160 +  return generate_fast_get_int_field0(T_BYTE);
   1.161 +}
   1.162 +
   1.163 +address JNI_FastGetField::generate_fast_get_char_field() {
   1.164 +  return generate_fast_get_int_field0(T_CHAR);
   1.165 +}
   1.166 +
   1.167 +address JNI_FastGetField::generate_fast_get_short_field() {
   1.168 +  return generate_fast_get_int_field0(T_SHORT);
   1.169 +}
   1.170 +
   1.171 +address JNI_FastGetField::generate_fast_get_int_field() {
   1.172 +  return generate_fast_get_int_field0(T_INT);
   1.173 +}
   1.174 +
   1.175 +address JNI_FastGetField::generate_fast_get_long_field() {
   1.176 +  const char *name = "jni_fast_GetLongField";
   1.177 +  ResourceMark rm;
   1.178 +  BufferBlob* blob = BufferBlob::create(name, BUFFER_SIZE*wordSize);
   1.179 +  CodeBuffer cbuf(blob);
   1.180 +  MacroAssembler* masm = new MacroAssembler(&cbuf);
   1.181 +  address fast_entry = __ pc();
   1.182 +
   1.183 +  Label slow;
   1.184 +
   1.185 +  // stack layout:    offset from rsp (in words):
   1.186 +  //  old rsi          0
   1.187 +  //  return pc        1
   1.188 +  //  jni env          2
   1.189 +  //  obj              3
   1.190 +  //  jfieldID         4
   1.191 +
   1.192 +  ExternalAddress counter(SafepointSynchronize::safepoint_counter_addr());
   1.193 +
   1.194 +  __ push  (rsi);
   1.195 +  __ mov32 (rcx, counter);
   1.196 +  __ testb (rcx, 1);
   1.197 +  __ jcc (Assembler::notZero, slow);
   1.198 +  if (os::is_MP()) {
   1.199 +    __ mov(rax, rcx);
   1.200 +    __ andptr(rax, 1);                         // rax, must end up 0
   1.201 +    __ movptr(rdx, Address(rsp, rax, Address::times_1, 3*wordSize));
   1.202 +                                              // obj, notice rax, is 0.
   1.203 +                                              // rdx is data dependent on rcx.
   1.204 +  } else {
   1.205 +    __ movptr(rdx, Address(rsp, 3*wordSize));  // obj
   1.206 +  }
   1.207 +  __ movptr(rsi, Address(rsp, 4*wordSize));  // jfieldID
   1.208 +  __ movptr(rdx, Address(rdx, 0));           // *obj
   1.209 +  __ shrptr(rsi, 2);                         // offset
   1.210 +
   1.211 +  assert(count < LIST_CAPACITY-1, "LIST_CAPACITY too small");
   1.212 +  speculative_load_pclist[count++] = __ pc();
   1.213 +  __ movptr(rax, Address(rdx, rsi, Address::times_1));
   1.214 +#ifndef _LP64
   1.215 +  speculative_load_pclist[count] = __ pc();
   1.216 +  __ movl(rdx, Address(rdx, rsi, Address::times_1, 4));
   1.217 +#endif // _LP64
   1.218 +
   1.219 +  if (os::is_MP()) {
   1.220 +    __ lea(rsi, counter);
   1.221 +    __ xorptr(rsi, rdx);
   1.222 +    __ xorptr(rsi, rax);
   1.223 +    __ xorptr(rsi, rdx);
   1.224 +    __ xorptr(rsi, rax);
   1.225 +    __ cmp32(rcx, Address(rsi, 0));
   1.226 +    // ca1 is the same as ca because
   1.227 +    // rax, ^ rdx ^ counter_addr ^ rax, ^ rdx = address
   1.228 +    // ca1 is data dependent on both rax, and rdx.
   1.229 +  } else {
   1.230 +    __ cmp32(rcx, counter);
   1.231 +  }
   1.232 +  __ jcc (Assembler::notEqual, slow);
   1.233 +
   1.234 +  __ pop (rsi);
   1.235 +
   1.236 +#ifndef _WINDOWS
   1.237 +  __ ret (0);
   1.238 +#else
   1.239 +  // __stdcall calling convention
   1.240 +  __ ret (3*wordSize);
   1.241 +#endif
   1.242 +
   1.243 +  slowcase_entry_pclist[count-1] = __ pc();
   1.244 +  slowcase_entry_pclist[count++] = __ pc();
   1.245 +  __ bind (slow);
   1.246 +  __ pop  (rsi);
   1.247 +  address slow_case_addr = jni_GetLongField_addr();;
   1.248 +  // tail call
   1.249 +  __ jump (ExternalAddress(slow_case_addr));
   1.250 +
   1.251 +  __ flush ();
   1.252 +
   1.253 +#ifndef _WINDOWS
   1.254 +  return fast_entry;
   1.255 +#else
   1.256 +  jni_fast_GetLongField_fp = (GetLongField_t) fast_entry;
   1.257 +  return os::win32::fast_jni_accessor_wrapper(T_LONG);
   1.258 +#endif
   1.259 +}
   1.260 +
   1.261 +address JNI_FastGetField::generate_fast_get_float_field0(BasicType type) {
   1.262 +  const char *name;
   1.263 +  switch (type) {
   1.264 +    case T_FLOAT:  name = "jni_fast_GetFloatField";  break;
   1.265 +    case T_DOUBLE: name = "jni_fast_GetDoubleField"; break;
   1.266 +    default:       ShouldNotReachHere();
   1.267 +  }
   1.268 +  ResourceMark rm;
   1.269 +  BufferBlob* blob = BufferBlob::create(name, BUFFER_SIZE*wordSize);
   1.270 +  CodeBuffer cbuf(blob);
   1.271 +  MacroAssembler* masm = new MacroAssembler(&cbuf);
   1.272 +  address fast_entry = __ pc();
   1.273 +
   1.274 +  Label slow_with_pop, slow;
   1.275 +
   1.276 +  // stack layout:    offset from rsp (in words):
   1.277 +  //  return pc        0
   1.278 +  //  jni env          1
   1.279 +  //  obj              2
   1.280 +  //  jfieldID         3
   1.281 +
   1.282 +  ExternalAddress counter(SafepointSynchronize::safepoint_counter_addr());
   1.283 +
   1.284 +  __ mov32 (rcx, counter);
   1.285 +  __ testb (rcx, 1);
   1.286 +  __ jcc (Assembler::notZero, slow);
   1.287 +  if (os::is_MP()) {
   1.288 +    __ mov(rax, rcx);
   1.289 +    __ andptr(rax, 1);                         // rax, must end up 0
   1.290 +    __ movptr(rdx, Address(rsp, rax, Address::times_1, 2*wordSize));
   1.291 +                                              // obj, notice rax, is 0.
   1.292 +                                              // rdx is data dependent on rcx.
   1.293 +  } else {
   1.294 +    __ movptr(rdx, Address(rsp, 2*wordSize)); // obj
   1.295 +  }
   1.296 +  __ movptr(rax, Address(rsp, 3*wordSize));  // jfieldID
   1.297 +  __ movptr(rdx, Address(rdx, 0));           // *obj
   1.298 +  __ shrptr(rax, 2);                         // offset
   1.299 +
   1.300 +  assert(count < LIST_CAPACITY, "LIST_CAPACITY too small");
   1.301 +  speculative_load_pclist[count] = __ pc();
   1.302 +  switch (type) {
   1.303 +#ifndef _LP64
   1.304 +    case T_FLOAT:  __ fld_s (Address(rdx, rax, Address::times_1)); break;
   1.305 +    case T_DOUBLE: __ fld_d (Address(rdx, rax, Address::times_1)); break;
   1.306 +#else
   1.307 +    case T_FLOAT:  __ movflt (xmm0, Address(robj, roffset, Address::times_1)); break;
   1.308 +    case T_DOUBLE: __ movdbl (xmm0, Address(robj, roffset, Address::times_1)); break;
   1.309 +#endif // _LP64
   1.310 +    default:       ShouldNotReachHere();
   1.311 +  }
   1.312 +
   1.313 +  Address ca1;
   1.314 +  if (os::is_MP()) {
   1.315 +    __ fst_s (Address(rsp, -4));
   1.316 +    __ lea(rdx, counter);
   1.317 +    __ movl (rax, Address(rsp, -4));
   1.318 +    // garbage hi-order bits on 64bit are harmless.
   1.319 +    __ xorptr(rdx, rax);
   1.320 +    __ xorptr(rdx, rax);
   1.321 +    __ cmp32(rcx, Address(rdx, 0));
   1.322 +                                          // rax, ^ counter_addr ^ rax, = address
   1.323 +                                          // ca1 is data dependent on the field
   1.324 +                                          // access.
   1.325 +  } else {
   1.326 +    __ cmp32(rcx, counter);
   1.327 +  }
   1.328 +  __ jcc (Assembler::notEqual, slow_with_pop);
   1.329 +
   1.330 +#ifndef _WINDOWS
   1.331 +  __ ret (0);
   1.332 +#else
   1.333 +  // __stdcall calling convention
   1.334 +  __ ret (3*wordSize);
   1.335 +#endif
   1.336 +
   1.337 +  __ bind (slow_with_pop);
   1.338 +  // invalid load. pop FPU stack.
   1.339 +  __ fstp_d (0);
   1.340 +
   1.341 +  slowcase_entry_pclist[count++] = __ pc();
   1.342 +  __ bind (slow);
   1.343 +  address slow_case_addr;
   1.344 +  switch (type) {
   1.345 +    case T_FLOAT:  slow_case_addr = jni_GetFloatField_addr();  break;
   1.346 +    case T_DOUBLE: slow_case_addr = jni_GetDoubleField_addr(); break;
   1.347 +    default:       ShouldNotReachHere();
   1.348 +  }
   1.349 +  // tail call
   1.350 +  __ jump (ExternalAddress(slow_case_addr));
   1.351 +
   1.352 +  __ flush ();
   1.353 +
   1.354 +#ifndef _WINDOWS
   1.355 +  return fast_entry;
   1.356 +#else
   1.357 +  switch (type) {
   1.358 +  case T_FLOAT:  jni_fast_GetFloatField_fp  = (GetFloatField_t)  fast_entry; break;
   1.359 +  case T_DOUBLE: jni_fast_GetDoubleField_fp = (GetDoubleField_t) fast_entry; break;
   1.360 +  }
   1.361 +  return os::win32::fast_jni_accessor_wrapper(type);
   1.362 +#endif
   1.363 +}
   1.364 +
   1.365 +address JNI_FastGetField::generate_fast_get_float_field() {
   1.366 +  return generate_fast_get_float_field0(T_FLOAT);
   1.367 +}
   1.368 +
   1.369 +address JNI_FastGetField::generate_fast_get_double_field() {
   1.370 +  return generate_fast_get_float_field0(T_DOUBLE);
   1.371 +}

mercurial