src/os_cpu/aix_ppc/vm/atomic_aix_ppc.inline.hpp

changeset 6465
666e6ce3976c
child 6501
c668f307a4c0
     1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/src/os_cpu/aix_ppc/vm/atomic_aix_ppc.inline.hpp	Fri Sep 06 20:16:09 2013 +0200
     1.3 @@ -0,0 +1,401 @@
     1.4 +/*
     1.5 + * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
     1.6 + * Copyright 2012, 2013 SAP AG. All rights reserved.
     1.7 + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
     1.8 + *
     1.9 + * This code is free software; you can redistribute it and/or modify it
    1.10 + * under the terms of the GNU General Public License version 2 only, as
    1.11 + * published by the Free Software Foundation.
    1.12 + *
    1.13 + * This code is distributed in the hope that it will be useful, but WITHOUT
    1.14 + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
    1.15 + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
    1.16 + * version 2 for more details (a copy is included in the LICENSE file that
    1.17 + * accompanied this code).
    1.18 + *
    1.19 + * You should have received a copy of the GNU General Public License version
    1.20 + * 2 along with this work; if not, write to the Free Software Foundation,
    1.21 + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
    1.22 + *
    1.23 + * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
    1.24 + * or visit www.oracle.com if you need additional information or have any
    1.25 + * questions.
    1.26 + *
    1.27 + */
    1.28 +
    1.29 +#ifndef OS_CPU_AIX_OJDKPPC_VM_ATOMIC_AIX_PPC_INLINE_HPP
    1.30 +#define OS_CPU_AIX_OJDKPPC_VM_ATOMIC_AIX_PPC_INLINE_HPP
    1.31 +
    1.32 +#include "orderAccess_aix_ppc.inline.hpp"
    1.33 +#include "runtime/atomic.hpp"
    1.34 +#include "runtime/os.hpp"
    1.35 +#include "vm_version_ppc.hpp"
    1.36 +
    1.37 +#ifndef _LP64
    1.38 +#error "Atomic currently only impleneted for PPC64"
    1.39 +#endif
    1.40 +
    1.41 +// Implementation of class atomic
    1.42 +
    1.43 +inline void Atomic::store    (jbyte    store_value, jbyte*    dest) { *dest = store_value; }
    1.44 +inline void Atomic::store    (jshort   store_value, jshort*   dest) { *dest = store_value; }
    1.45 +inline void Atomic::store    (jint     store_value, jint*     dest) { *dest = store_value; }
    1.46 +inline void Atomic::store    (jlong    store_value, jlong*    dest) { *dest = store_value; }
    1.47 +inline void Atomic::store_ptr(intptr_t store_value, intptr_t* dest) { *dest = store_value; }
    1.48 +inline void Atomic::store_ptr(void*    store_value, void*     dest) { *(void**)dest = store_value; }
    1.49 +
    1.50 +inline void Atomic::store    (jbyte    store_value, volatile jbyte*    dest) { *dest = store_value; }
    1.51 +inline void Atomic::store    (jshort   store_value, volatile jshort*   dest) { *dest = store_value; }
    1.52 +inline void Atomic::store    (jint     store_value, volatile jint*     dest) { *dest = store_value; }
    1.53 +inline void Atomic::store    (jlong    store_value, volatile jlong*    dest) { *dest = store_value; }
    1.54 +inline void Atomic::store_ptr(intptr_t store_value, volatile intptr_t* dest) { *dest = store_value; }
    1.55 +inline void Atomic::store_ptr(void*    store_value, volatile void*     dest) { *(void* volatile *)dest = store_value; }
    1.56 +
    1.57 +inline jlong Atomic::load(volatile jlong* src) { return *src; }
    1.58 +
    1.59 +/*
    1.60 +  machine barrier instructions:
    1.61 +
    1.62 +  - ppc_sync            two-way memory barrier, aka fence
    1.63 +  - ppc_lwsync          orders  Store|Store,
    1.64 +                                 Load|Store,
    1.65 +                                 Load|Load,
    1.66 +                        but not Store|Load
    1.67 +  - ppc_eieio           orders memory accesses for device memory (only)
    1.68 +  - ppc_isync           invalidates speculatively executed instructions
    1.69 +                        From the POWER ISA 2.06 documentation:
    1.70 +                         "[...] an isync instruction prevents the execution of
    1.71 +                        instructions following the isync until instructions
    1.72 +                        preceding the isync have completed, [...]"
    1.73 +                        From IBM's AIX assembler reference:
    1.74 +                         "The isync [...] instructions causes the processor to
    1.75 +                        refetch any instructions that might have been fetched
    1.76 +                        prior to the isync instruction. The instruction isync
    1.77 +                        causes the processor to wait for all previous instructions
    1.78 +                        to complete. Then any instructions already fetched are
    1.79 +                        discarded and instruction processing continues in the
    1.80 +                        environment established by the previous instructions."
    1.81 +
    1.82 +  semantic barrier instructions:
    1.83 +  (as defined in orderAccess.hpp)
    1.84 +
    1.85 +  - ppc_release         orders Store|Store,       (maps to ppc_lwsync)
    1.86 +                                Load|Store
    1.87 +  - ppc_acquire         orders  Load|Store,       (maps to ppc_lwsync)
    1.88 +                                Load|Load
    1.89 +  - ppc_fence           orders Store|Store,       (maps to ppc_sync)
    1.90 +                                Load|Store,
    1.91 +                                Load|Load,
    1.92 +                               Store|Load
    1.93 +*/
    1.94 +
    1.95 +#define strasm_ppc_sync                       "\n  sync    \n"
    1.96 +#define strasm_ppc_lwsync                     "\n  lwsync  \n"
    1.97 +#define strasm_ppc_isync                      "\n  isync   \n"
    1.98 +#define strasm_ppc_release                    strasm_ppc_lwsync
    1.99 +#define strasm_ppc_acquire                    strasm_ppc_lwsync
   1.100 +#define strasm_ppc_fence                      strasm_ppc_sync
   1.101 +#define strasm_ppc_nobarrier                  ""
   1.102 +#define strasm_ppc_nobarrier_clobber_memory   ""
   1.103 +
   1.104 +inline jint     Atomic::add    (jint     add_value, volatile jint*     dest) {
   1.105 +
   1.106 +  unsigned int result;
   1.107 +
   1.108 +  __asm__ __volatile__ (
   1.109 +    strasm_ppc_lwsync
   1.110 +    "1: lwarx   %0,  0, %2    \n"
   1.111 +    "   add     %0, %0, %1    \n"
   1.112 +    "   stwcx.  %0,  0, %2    \n"
   1.113 +    "   bne-    1b            \n"
   1.114 +    strasm_ppc_isync
   1.115 +    : /*%0*/"=&r" (result)
   1.116 +    : /*%1*/"r" (add_value), /*%2*/"r" (dest)
   1.117 +    : "cc", "memory" );
   1.118 +
   1.119 +  return (jint) result;
   1.120 +}
   1.121 +
   1.122 +
   1.123 +inline intptr_t Atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest) {
   1.124 +
   1.125 +  long result;
   1.126 +
   1.127 +  __asm__ __volatile__ (
   1.128 +    strasm_ppc_lwsync
   1.129 +    "1: ldarx   %0,  0, %2    \n"
   1.130 +    "   add     %0, %0, %1    \n"
   1.131 +    "   stdcx.  %0,  0, %2    \n"
   1.132 +    "   bne-    1b            \n"
   1.133 +    strasm_ppc_isync
   1.134 +    : /*%0*/"=&r" (result)
   1.135 +    : /*%1*/"r" (add_value), /*%2*/"r" (dest)
   1.136 +    : "cc", "memory" );
   1.137 +
   1.138 +  return (intptr_t) result;
   1.139 +}
   1.140 +
   1.141 +inline void*    Atomic::add_ptr(intptr_t add_value, volatile void*     dest) {
   1.142 +  return (void*)add_ptr(add_value, (volatile intptr_t*)dest);
   1.143 +}
   1.144 +
   1.145 +
   1.146 +inline void Atomic::inc    (volatile jint*     dest) {
   1.147 +
   1.148 +  unsigned int temp;
   1.149 +
   1.150 +  __asm__ __volatile__ (
   1.151 +    strasm_ppc_nobarrier
   1.152 +    "1: lwarx   %0,  0, %2    \n"
   1.153 +    "   addic   %0, %0,  1    \n"
   1.154 +    "   stwcx.  %0,  0, %2    \n"
   1.155 +    "   bne-    1b            \n"
   1.156 +    strasm_ppc_nobarrier
   1.157 +    : /*%0*/"=&r" (temp), "=m" (*dest)
   1.158 +    : /*%2*/"r" (dest), "m" (*dest)
   1.159 +    : "cc" strasm_ppc_nobarrier_clobber_memory);
   1.160 +
   1.161 +}
   1.162 +
   1.163 +inline void Atomic::inc_ptr(volatile intptr_t* dest) {
   1.164 +
   1.165 +  long temp;
   1.166 +
   1.167 +  __asm__ __volatile__ (
   1.168 +    strasm_ppc_nobarrier
   1.169 +    "1: ldarx   %0,  0, %2    \n"
   1.170 +    "   addic   %0, %0,  1    \n"
   1.171 +    "   stdcx.  %0,  0, %2    \n"
   1.172 +    "   bne-    1b            \n"
   1.173 +    strasm_ppc_nobarrier
   1.174 +    : /*%0*/"=&r" (temp), "=m" (*dest)
   1.175 +    : /*%2*/"r" (dest), "m" (*dest)
   1.176 +    : "cc" strasm_ppc_nobarrier_clobber_memory);
   1.177 +
   1.178 +}
   1.179 +
   1.180 +inline void Atomic::inc_ptr(volatile void*     dest) {
   1.181 +  inc_ptr((volatile intptr_t*)dest);
   1.182 +}
   1.183 +
   1.184 +
   1.185 +inline void Atomic::dec    (volatile jint*     dest) {
   1.186 +
   1.187 +  unsigned int temp;
   1.188 +
   1.189 +  __asm__ __volatile__ (
   1.190 +    strasm_ppc_nobarrier
   1.191 +    "1: lwarx   %0,  0, %2    \n"
   1.192 +    "   addic   %0, %0, -1    \n"
   1.193 +    "   stwcx.  %0,  0, %2    \n"
   1.194 +    "   bne-    1b            \n"
   1.195 +    strasm_ppc_nobarrier
   1.196 +    : /*%0*/"=&r" (temp), "=m" (*dest)
   1.197 +    : /*%2*/"r" (dest), "m" (*dest)
   1.198 +    : "cc" strasm_ppc_nobarrier_clobber_memory);
   1.199 +
   1.200 +}
   1.201 +
   1.202 +inline void Atomic::dec_ptr(volatile intptr_t* dest) {
   1.203 +
   1.204 +  long temp;
   1.205 +
   1.206 +  __asm__ __volatile__ (
   1.207 +    strasm_ppc_nobarrier
   1.208 +    "1: ldarx   %0,  0, %2    \n"
   1.209 +    "   addic   %0, %0, -1    \n"
   1.210 +    "   stdcx.  %0,  0, %2    \n"
   1.211 +    "   bne-    1b            \n"
   1.212 +    strasm_ppc_nobarrier
   1.213 +    : /*%0*/"=&r" (temp), "=m" (*dest)
   1.214 +    : /*%2*/"r" (dest), "m" (*dest)
   1.215 +    : "cc" strasm_ppc_nobarrier_clobber_memory);
   1.216 +
   1.217 +}
   1.218 +
   1.219 +inline void Atomic::dec_ptr(volatile void*     dest) {
   1.220 +  dec_ptr((volatile intptr_t*)dest);
   1.221 +}
   1.222 +
   1.223 +inline jint Atomic::xchg(jint exchange_value, volatile jint* dest) {
   1.224 +
   1.225 +  // Note that xchg_ptr doesn't necessarily do an acquire
   1.226 +  // (see synchronizer.cpp).
   1.227 +
   1.228 +  unsigned int old_value;
   1.229 +  const uint64_t zero = 0;
   1.230 +
   1.231 +  __asm__ __volatile__ (
   1.232 +    /* lwsync */
   1.233 +    strasm_ppc_lwsync
   1.234 +    /* atomic loop */
   1.235 +    "1:                                                 \n"
   1.236 +    "   lwarx   %[old_value], %[dest], %[zero]          \n"
   1.237 +    "   stwcx.  %[exchange_value], %[dest], %[zero]     \n"
   1.238 +    "   bne-    1b                                      \n"
   1.239 +    /* isync */
   1.240 +    strasm_ppc_sync
   1.241 +    /* exit */
   1.242 +    "2:                                                 \n"
   1.243 +    /* out */
   1.244 +    : [old_value]       "=&r"   (old_value),
   1.245 +                        "=m"    (*dest)
   1.246 +    /* in */
   1.247 +    : [dest]            "b"     (dest),
   1.248 +      [zero]            "r"     (zero),
   1.249 +      [exchange_value]  "r"     (exchange_value),
   1.250 +                        "m"     (*dest)
   1.251 +    /* clobber */
   1.252 +    : "cc",
   1.253 +      "memory"
   1.254 +    );
   1.255 +
   1.256 +  return (jint) old_value;
   1.257 +}
   1.258 +
   1.259 +inline intptr_t Atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest) {
   1.260 +
   1.261 +  // Note that xchg_ptr doesn't necessarily do an acquire
   1.262 +  // (see synchronizer.cpp).
   1.263 +
   1.264 +  long old_value;
   1.265 +  const uint64_t zero = 0;
   1.266 +
   1.267 +  __asm__ __volatile__ (
   1.268 +    /* lwsync */
   1.269 +    strasm_ppc_lwsync
   1.270 +    /* atomic loop */
   1.271 +    "1:                                                 \n"
   1.272 +    "   ldarx   %[old_value], %[dest], %[zero]          \n"
   1.273 +    "   stdcx.  %[exchange_value], %[dest], %[zero]     \n"
   1.274 +    "   bne-    1b                                      \n"
   1.275 +    /* isync */
   1.276 +    strasm_ppc_sync
   1.277 +    /* exit */
   1.278 +    "2:                                                 \n"
   1.279 +    /* out */
   1.280 +    : [old_value]       "=&r"   (old_value),
   1.281 +                        "=m"    (*dest)
   1.282 +    /* in */
   1.283 +    : [dest]            "b"     (dest),
   1.284 +      [zero]            "r"     (zero),
   1.285 +      [exchange_value]  "r"     (exchange_value),
   1.286 +                        "m"     (*dest)
   1.287 +    /* clobber */
   1.288 +    : "cc",
   1.289 +      "memory"
   1.290 +    );
   1.291 +
   1.292 +  return (intptr_t) old_value;
   1.293 +}
   1.294 +
   1.295 +inline void* Atomic::xchg_ptr(void* exchange_value, volatile void* dest) {
   1.296 +  return (void*)xchg_ptr((intptr_t)exchange_value, (volatile intptr_t*)dest);
   1.297 +}
   1.298 +
   1.299 +inline jint Atomic::cmpxchg(jint exchange_value, volatile jint* dest, jint compare_value) {
   1.300 +
   1.301 +  // Note that cmpxchg guarantees a two-way memory barrier across
   1.302 +  // the cmpxchg, so it's really a a 'fence_cmpxchg_acquire'
   1.303 +  // (see atomic.hpp).
   1.304 +
   1.305 +  unsigned int old_value;
   1.306 +  const uint64_t zero = 0;
   1.307 +
   1.308 +  __asm__ __volatile__ (
   1.309 +    /* fence */
   1.310 +    strasm_ppc_sync
   1.311 +    /* simple guard */
   1.312 +    "   lwz     %[old_value], 0(%[dest])                \n"
   1.313 +    "   cmpw    %[compare_value], %[old_value]          \n"
   1.314 +    "   bne-    2f                                      \n"
   1.315 +    /* atomic loop */
   1.316 +    "1:                                                 \n"
   1.317 +    "   lwarx   %[old_value], %[dest], %[zero]          \n"
   1.318 +    "   cmpw    %[compare_value], %[old_value]          \n"
   1.319 +    "   bne-    2f                                      \n"
   1.320 +    "   stwcx.  %[exchange_value], %[dest], %[zero]     \n"
   1.321 +    "   bne-    1b                                      \n"
   1.322 +    /* acquire */
   1.323 +    strasm_ppc_sync
   1.324 +    /* exit */
   1.325 +    "2:                                                 \n"
   1.326 +    /* out */
   1.327 +    : [old_value]       "=&r"   (old_value),
   1.328 +                        "=m"    (*dest)
   1.329 +    /* in */
   1.330 +    : [dest]            "b"     (dest),
   1.331 +      [zero]            "r"     (zero),
   1.332 +      [compare_value]   "r"     (compare_value),
   1.333 +      [exchange_value]  "r"     (exchange_value),
   1.334 +                        "m"     (*dest)
   1.335 +    /* clobber */
   1.336 +    : "cc",
   1.337 +      "memory"
   1.338 +    );
   1.339 +
   1.340 +  return (jint) old_value;
   1.341 +}
   1.342 +
   1.343 +inline jlong Atomic::cmpxchg(jlong exchange_value, volatile jlong* dest, jlong compare_value) {
   1.344 +
   1.345 +  // Note that cmpxchg guarantees a two-way memory barrier across
   1.346 +  // the cmpxchg, so it's really a a 'fence_cmpxchg_acquire'
   1.347 +  // (see atomic.hpp).
   1.348 +
   1.349 +  long old_value;
   1.350 +  const uint64_t zero = 0;
   1.351 +
   1.352 +  __asm__ __volatile__ (
   1.353 +    /* fence */
   1.354 +    strasm_ppc_sync
   1.355 +    /* simple guard */
   1.356 +    "   ld      %[old_value], 0(%[dest])                \n"
   1.357 +    "   cmpd    %[compare_value], %[old_value]          \n"
   1.358 +    "   bne-    2f                                      \n"
   1.359 +    /* atomic loop */
   1.360 +    "1:                                                 \n"
   1.361 +    "   ldarx   %[old_value], %[dest], %[zero]          \n"
   1.362 +    "   cmpd    %[compare_value], %[old_value]          \n"
   1.363 +    "   bne-    2f                                      \n"
   1.364 +    "   stdcx.  %[exchange_value], %[dest], %[zero]     \n"
   1.365 +    "   bne-    1b                                      \n"
   1.366 +    /* acquire */
   1.367 +    strasm_ppc_sync
   1.368 +    /* exit */
   1.369 +    "2:                                                 \n"
   1.370 +    /* out */
   1.371 +    : [old_value]       "=&r"   (old_value),
   1.372 +                        "=m"    (*dest)
   1.373 +    /* in */
   1.374 +    : [dest]            "b"     (dest),
   1.375 +      [zero]            "r"     (zero),
   1.376 +      [compare_value]   "r"     (compare_value),
   1.377 +      [exchange_value]  "r"     (exchange_value),
   1.378 +                        "m"     (*dest)
   1.379 +    /* clobber */
   1.380 +    : "cc",
   1.381 +      "memory"
   1.382 +    );
   1.383 +
   1.384 +  return (jlong) old_value;
   1.385 +}
   1.386 +
   1.387 +inline intptr_t Atomic::cmpxchg_ptr(intptr_t exchange_value, volatile intptr_t* dest, intptr_t compare_value) {
   1.388 +  return (intptr_t)cmpxchg((jlong)exchange_value, (volatile jlong*)dest, (jlong)compare_value);
   1.389 +}
   1.390 +
   1.391 +inline void* Atomic::cmpxchg_ptr(void* exchange_value, volatile void* dest, void* compare_value) {
   1.392 +  return (void*)cmpxchg((jlong)exchange_value, (volatile jlong*)dest, (jlong)compare_value);
   1.393 +}
   1.394 +
   1.395 +#undef strasm_ppc_sync
   1.396 +#undef strasm_ppc_lwsync
   1.397 +#undef strasm_ppc_isync
   1.398 +#undef strasm_ppc_release
   1.399 +#undef strasm_ppc_acquire
   1.400 +#undef strasm_ppc_fence
   1.401 +#undef strasm_ppc_nobarrier
   1.402 +#undef strasm_ppc_nobarrier_clobber_memory
   1.403 +
   1.404 +#endif // OS_CPU_AIX_OJDKPPC_VM_ATOMIC_AIX_PPC_INLINE_HPP

mercurial