src/cpu/sparc/vm/vm_version_sparc.cpp

Fri, 07 Jun 2013 16:46:37 -0700

author
morris
date
Fri, 07 Jun 2013 16:46:37 -0700
changeset 5283
46c544b8fbfc
parent 4430
4a916f2ce331
child 6154
eae426d683f6
permissions
-rw-r--r--

8008407: remove SPARC V8 support
Summary: Removed most of the SPARC V8 instructions
Reviewed-by: kvn, twisti

duke@435 1 /*
kvn@3882 2 * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved.
duke@435 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435 4 *
duke@435 5 * This code is free software; you can redistribute it and/or modify it
duke@435 6 * under the terms of the GNU General Public License version 2 only, as
duke@435 7 * published by the Free Software Foundation.
duke@435 8 *
duke@435 9 * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
duke@435 12 * version 2 for more details (a copy is included in the LICENSE file that
duke@435 13 * accompanied this code).
duke@435 14 *
duke@435 15 * You should have received a copy of the GNU General Public License version
duke@435 16 * 2 along with this work; if not, write to the Free Software Foundation,
duke@435 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435 18 *
trims@1907 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
trims@1907 20 * or visit www.oracle.com if you need additional information or have any
trims@1907 21 * questions.
duke@435 22 *
duke@435 23 */
duke@435 24
stefank@2314 25 #include "precompiled.hpp"
twisti@4323 26 #include "asm/macroAssembler.inline.hpp"
stefank@2314 27 #include "memory/resourceArea.hpp"
stefank@2314 28 #include "runtime/java.hpp"
stefank@2314 29 #include "runtime/stubCodeGenerator.hpp"
stefank@2314 30 #include "vm_version_sparc.hpp"
stefank@2314 31 #ifdef TARGET_OS_FAMILY_linux
stefank@2314 32 # include "os_linux.inline.hpp"
stefank@2314 33 #endif
stefank@2314 34 #ifdef TARGET_OS_FAMILY_solaris
stefank@2314 35 # include "os_solaris.inline.hpp"
stefank@2314 36 #endif
duke@435 37
duke@435 38 int VM_Version::_features = VM_Version::unknown_m;
duke@435 39 const char* VM_Version::_features_str = "";
duke@435 40
duke@435 41 void VM_Version::initialize() {
duke@435 42 _features = determine_features();
duke@435 43 PrefetchCopyIntervalInBytes = prefetch_copy_interval_in_bytes();
duke@435 44 PrefetchScanIntervalInBytes = prefetch_scan_interval_in_bytes();
duke@435 45 PrefetchFieldsAhead = prefetch_fields_ahead();
duke@435 46
kvn@3052 47 assert(0 <= AllocatePrefetchInstr && AllocatePrefetchInstr <= 1, "invalid value");
kvn@3052 48 if( AllocatePrefetchInstr < 0 ) AllocatePrefetchInstr = 0;
kvn@3052 49 if( AllocatePrefetchInstr > 1 ) AllocatePrefetchInstr = 0;
kvn@3052 50
duke@435 51 // Allocation prefetch settings
kvn@3052 52 intx cache_line_size = prefetch_data_size();
duke@435 53 if( cache_line_size > AllocatePrefetchStepSize )
duke@435 54 AllocatePrefetchStepSize = cache_line_size;
kvn@3052 55
kvn@3052 56 assert(AllocatePrefetchLines > 0, "invalid value");
kvn@3052 57 if( AllocatePrefetchLines < 1 ) // set valid value in product VM
kvn@3052 58 AllocatePrefetchLines = 3;
kvn@3052 59 assert(AllocateInstancePrefetchLines > 0, "invalid value");
kvn@3052 60 if( AllocateInstancePrefetchLines < 1 ) // set valid value in product VM
kvn@3052 61 AllocateInstancePrefetchLines = 1;
duke@435 62
duke@435 63 AllocatePrefetchDistance = allocate_prefetch_distance();
duke@435 64 AllocatePrefetchStyle = allocate_prefetch_style();
duke@435 65
kvn@3052 66 assert((AllocatePrefetchDistance % AllocatePrefetchStepSize) == 0 &&
kvn@3052 67 (AllocatePrefetchDistance > 0), "invalid value");
kvn@3052 68 if ((AllocatePrefetchDistance % AllocatePrefetchStepSize) != 0 ||
kvn@3052 69 (AllocatePrefetchDistance <= 0)) {
kvn@3052 70 AllocatePrefetchDistance = AllocatePrefetchStepSize;
kvn@3052 71 }
duke@435 72
kvn@3037 73 if (AllocatePrefetchStyle == 3 && !has_blk_init()) {
kvn@3037 74 warning("BIS instructions are not available on this CPU");
kvn@3037 75 FLAG_SET_DEFAULT(AllocatePrefetchStyle, 1);
kvn@3037 76 }
kvn@3037 77
morris@5283 78 guarantee(VM_Version::has_v9(), "only SPARC v9 is supported");
morris@5283 79
morris@5283 80 assert(ArraycopySrcPrefetchDistance < 4096, "invalid value");
morris@5283 81 if (ArraycopySrcPrefetchDistance >= 4096)
morris@5283 82 ArraycopySrcPrefetchDistance = 4064;
morris@5283 83 assert(ArraycopyDstPrefetchDistance < 4096, "invalid value");
morris@5283 84 if (ArraycopyDstPrefetchDistance >= 4096)
morris@5283 85 ArraycopyDstPrefetchDistance = 4064;
kvn@3103 86
duke@435 87 UseSSE = 0; // Only on x86 and x64
duke@435 88
kvn@3052 89 _supports_cx8 = has_v9();
roland@4106 90 _supports_atomic_getset4 = true; // swap instruction
duke@435 91
kvn@2403 92 if (is_niagara()) {
duke@435 93 // Indirect branch is the same cost as direct
duke@435 94 if (FLAG_IS_DEFAULT(UseInlineCaches)) {
kvn@1110 95 FLAG_SET_DEFAULT(UseInlineCaches, false);
duke@435 96 }
kvn@2403 97 // Align loops on a single instruction boundary.
kvn@2403 98 if (FLAG_IS_DEFAULT(OptoLoopAlignment)) {
kvn@2403 99 FLAG_SET_DEFAULT(OptoLoopAlignment, 4);
kvn@2403 100 }
johnc@3997 101 // When using CMS or G1, we cannot use memset() in BOT updates
johnc@3997 102 // because the sun4v/CMT version in libc_psr uses BIS which
johnc@3997 103 // exposes "phantom zeros" to concurrent readers. See 6948537.
johnc@3997 104 if (FLAG_IS_DEFAULT(UseMemSetInBOT) && (UseConcMarkSweepGC || UseG1GC)) {
kvn@2403 105 FLAG_SET_DEFAULT(UseMemSetInBOT, false);
kvn@2403 106 }
coleenp@548 107 #ifdef _LP64
kvn@1077 108 // 32-bit oops don't make sense for the 64-bit VM on sparc
kvn@1077 109 // since the 32-bit VM has the same registers and smaller objects.
kvn@1077 110 Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes);
roland@4159 111 Universe::set_narrow_klass_shift(LogKlassAlignmentInBytes);
coleenp@548 112 #endif // _LP64
duke@435 113 #ifdef COMPILER2
duke@435 114 // Indirect branch is the same cost as direct
duke@435 115 if (FLAG_IS_DEFAULT(UseJumpTables)) {
kvn@1110 116 FLAG_SET_DEFAULT(UseJumpTables, true);
duke@435 117 }
duke@435 118 // Single-issue, so entry and loop tops are
duke@435 119 // aligned on a single instruction boundary
duke@435 120 if (FLAG_IS_DEFAULT(InteriorEntryAlignment)) {
kvn@1110 121 FLAG_SET_DEFAULT(InteriorEntryAlignment, 4);
duke@435 122 }
kvn@2403 123 if (is_niagara_plus()) {
kvn@3052 124 if (has_blk_init() && UseTLAB &&
kvn@3052 125 FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
kvn@3052 126 // Use BIS instruction for TLAB allocation prefetch.
kvn@3052 127 FLAG_SET_ERGO(intx, AllocatePrefetchInstr, 1);
kvn@3052 128 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
kvn@3052 129 FLAG_SET_ERGO(intx, AllocatePrefetchStyle, 3);
kvn@3052 130 }
kvn@1802 131 if (FLAG_IS_DEFAULT(AllocatePrefetchDistance)) {
kvn@3052 132 // Use smaller prefetch distance with BIS
kvn@1802 133 FLAG_SET_DEFAULT(AllocatePrefetchDistance, 64);
kvn@1802 134 }
kvn@1802 135 }
kvn@3052 136 if (is_T4()) {
kvn@3052 137 // Double number of prefetched cache lines on T4
kvn@3052 138 // since L2 cache line size is smaller (32 bytes).
kvn@3052 139 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) {
kvn@3052 140 FLAG_SET_ERGO(intx, AllocatePrefetchLines, AllocatePrefetchLines*2);
kvn@3052 141 }
kvn@3052 142 if (FLAG_IS_DEFAULT(AllocateInstancePrefetchLines)) {
kvn@3052 143 FLAG_SET_ERGO(intx, AllocateInstancePrefetchLines, AllocateInstancePrefetchLines*2);
kvn@3052 144 }
kvn@3052 145 }
kvn@1802 146 if (AllocatePrefetchStyle != 3 && FLAG_IS_DEFAULT(AllocatePrefetchDistance)) {
kvn@1802 147 // Use different prefetch distance without BIS
kvn@1802 148 FLAG_SET_DEFAULT(AllocatePrefetchDistance, 256);
kvn@1802 149 }
kvn@3052 150 if (AllocatePrefetchInstr == 1) {
kvn@3052 151 // Need a space at the end of TLAB for BIS since it
kvn@3052 152 // will fault when accessing memory outside of heap.
kvn@3052 153
kvn@3052 154 // +1 for rounding up to next cache line, +1 to be safe
kvn@3052 155 int lines = AllocatePrefetchLines + 2;
kvn@3052 156 int step_size = AllocatePrefetchStepSize;
kvn@3052 157 int distance = AllocatePrefetchDistance;
kvn@3052 158 _reserve_for_allocation_prefetch = (distance + step_size*lines)/(int)HeapWordSize;
kvn@3052 159 }
duke@435 160 }
duke@435 161 #endif
duke@435 162 }
duke@435 163
twisti@1078 164 // Use hardware population count instruction if available.
twisti@1078 165 if (has_hardware_popc()) {
twisti@1078 166 if (FLAG_IS_DEFAULT(UsePopCountInstruction)) {
kvn@1110 167 FLAG_SET_DEFAULT(UsePopCountInstruction, true);
twisti@1078 168 }
kvn@3037 169 } else if (UsePopCountInstruction) {
kvn@3037 170 warning("POPC instruction is not available on this CPU");
kvn@3037 171 FLAG_SET_DEFAULT(UsePopCountInstruction, false);
kvn@3037 172 }
kvn@3037 173
kvn@3037 174 // T4 and newer Sparc cpus have new compare and branch instruction.
kvn@3037 175 if (has_cbcond()) {
kvn@3037 176 if (FLAG_IS_DEFAULT(UseCBCond)) {
kvn@3037 177 FLAG_SET_DEFAULT(UseCBCond, true);
kvn@3037 178 }
kvn@3037 179 } else if (UseCBCond) {
kvn@3037 180 warning("CBCOND instruction is not available on this CPU");
kvn@3037 181 FLAG_SET_DEFAULT(UseCBCond, false);
twisti@1078 182 }
twisti@1078 183
kvn@3092 184 assert(BlockZeroingLowLimit > 0, "invalid value");
kvn@3092 185 if (has_block_zeroing()) {
kvn@3092 186 if (FLAG_IS_DEFAULT(UseBlockZeroing)) {
kvn@3092 187 FLAG_SET_DEFAULT(UseBlockZeroing, true);
kvn@3092 188 }
kvn@3092 189 } else if (UseBlockZeroing) {
kvn@3092 190 warning("BIS zeroing instructions are not available on this CPU");
kvn@3092 191 FLAG_SET_DEFAULT(UseBlockZeroing, false);
kvn@3092 192 }
kvn@3092 193
kvn@3103 194 assert(BlockCopyLowLimit > 0, "invalid value");
kvn@3103 195 if (has_block_zeroing()) { // has_blk_init() && is_T4(): core's local L2 cache
kvn@3103 196 if (FLAG_IS_DEFAULT(UseBlockCopy)) {
kvn@3103 197 FLAG_SET_DEFAULT(UseBlockCopy, true);
kvn@3103 198 }
kvn@3103 199 } else if (UseBlockCopy) {
kvn@3103 200 warning("BIS instructions are not available or expensive on this CPU");
kvn@3103 201 FLAG_SET_DEFAULT(UseBlockCopy, false);
kvn@3103 202 }
kvn@3103 203
never@2085 204 #ifdef COMPILER2
kvn@3037 205 // T4 and newer Sparc cpus have fast RDPC.
kvn@3037 206 if (has_fast_rdpc() && FLAG_IS_DEFAULT(UseRDPCForConstantTableBase)) {
twisti@3249 207 FLAG_SET_DEFAULT(UseRDPCForConstantTableBase, true);
kvn@3037 208 }
kvn@3037 209
never@2085 210 // Currently not supported anywhere.
never@2085 211 FLAG_SET_DEFAULT(UseFPUForSpilling, false);
kvn@3049 212
kvn@3882 213 MaxVectorSize = 8;
kvn@3882 214
kvn@3049 215 assert((InteriorEntryAlignment % relocInfo::addr_unit()) == 0, "alignment is not a multiple of NOP size");
never@2085 216 #endif
never@2085 217
kvn@3049 218 assert((CodeEntryAlignment % relocInfo::addr_unit()) == 0, "alignment is not a multiple of NOP size");
kvn@3049 219 assert((OptoLoopAlignment % relocInfo::addr_unit()) == 0, "alignment is not a multiple of NOP size");
kvn@3049 220
duke@435 221 char buf[512];
kvn@3037 222 jio_snprintf(buf, sizeof(buf), "%s%s%s%s%s%s%s%s%s%s%s%s%s%s",
kvn@3037 223 (has_v9() ? ", v9" : (has_v8() ? ", v8" : "")),
twisti@1078 224 (has_hardware_popc() ? ", popc" : ""),
kvn@3037 225 (has_vis1() ? ", vis1" : ""),
kvn@3037 226 (has_vis2() ? ", vis2" : ""),
kvn@3037 227 (has_vis3() ? ", vis3" : ""),
kvn@3037 228 (has_blk_init() ? ", blk_init" : ""),
kvn@3037 229 (has_cbcond() ? ", cbcond" : ""),
kvn@3037 230 (is_ultra3() ? ", ultra3" : ""),
kvn@3037 231 (is_sun4v() ? ", sun4v" : ""),
kvn@3037 232 (is_niagara_plus() ? ", niagara_plus" : (is_niagara() ? ", niagara" : "")),
kvn@3037 233 (is_sparc64() ? ", sparc64" : ""),
twisti@1076 234 (!has_hardware_mul32() ? ", no-mul32" : ""),
twisti@1076 235 (!has_hardware_div32() ? ", no-div32" : ""),
duke@435 236 (!has_hardware_fsmuld() ? ", no-fsmuld" : ""));
duke@435 237
duke@435 238 // buf is started with ", " or is empty
duke@435 239 _features_str = strdup(strlen(buf) > 2 ? buf + 2 : buf);
duke@435 240
kvn@3001 241 // UseVIS is set to the smallest of what hardware supports and what
kvn@3001 242 // the command line requires. I.e., you cannot set UseVIS to 3 on
kvn@3001 243 // older UltraSparc which do not support it.
kvn@3001 244 if (UseVIS > 3) UseVIS=3;
kvn@3001 245 if (UseVIS < 0) UseVIS=0;
kvn@3001 246 if (!has_vis3()) // Drop to 2 if no VIS3 support
kvn@3001 247 UseVIS = MIN2((intx)2,UseVIS);
kvn@3001 248 if (!has_vis2()) // Drop to 1 if no VIS2 support
kvn@3001 249 UseVIS = MIN2((intx)1,UseVIS);
kvn@3001 250 if (!has_vis1()) // Drop to 0 if no VIS1 support
kvn@3001 251 UseVIS = 0;
kvn@3001 252
jwilhelm@4430 253 if (FLAG_IS_DEFAULT(ContendedPaddingWidth) &&
jwilhelm@4430 254 (cache_line_size > ContendedPaddingWidth))
jwilhelm@4430 255 ContendedPaddingWidth = cache_line_size;
jwilhelm@4430 256
duke@435 257 #ifndef PRODUCT
duke@435 258 if (PrintMiscellaneous && Verbose) {
kvn@3052 259 tty->print("Allocation");
duke@435 260 if (AllocatePrefetchStyle <= 0) {
kvn@3052 261 tty->print_cr(": no prefetching");
duke@435 262 } else {
kvn@3052 263 tty->print(" prefetching: ");
kvn@3052 264 if (AllocatePrefetchInstr == 0) {
kvn@3052 265 tty->print("PREFETCH");
kvn@3052 266 } else if (AllocatePrefetchInstr == 1) {
kvn@3052 267 tty->print("BIS");
kvn@3052 268 }
duke@435 269 if (AllocatePrefetchLines > 1) {
kvn@3052 270 tty->print_cr(" at distance %d, %d lines of %d bytes", AllocatePrefetchDistance, AllocatePrefetchLines, AllocatePrefetchStepSize);
duke@435 271 } else {
kvn@3052 272 tty->print_cr(" at distance %d, one line of %d bytes", AllocatePrefetchDistance, AllocatePrefetchStepSize);
duke@435 273 }
duke@435 274 }
duke@435 275 if (PrefetchCopyIntervalInBytes > 0) {
duke@435 276 tty->print_cr("PrefetchCopyIntervalInBytes %d", PrefetchCopyIntervalInBytes);
duke@435 277 }
duke@435 278 if (PrefetchScanIntervalInBytes > 0) {
duke@435 279 tty->print_cr("PrefetchScanIntervalInBytes %d", PrefetchScanIntervalInBytes);
duke@435 280 }
duke@435 281 if (PrefetchFieldsAhead > 0) {
duke@435 282 tty->print_cr("PrefetchFieldsAhead %d", PrefetchFieldsAhead);
duke@435 283 }
jwilhelm@4430 284 if (ContendedPaddingWidth > 0) {
jwilhelm@4430 285 tty->print_cr("ContendedPaddingWidth %d", ContendedPaddingWidth);
jwilhelm@4430 286 }
duke@435 287 }
duke@435 288 #endif // PRODUCT
duke@435 289 }
duke@435 290
duke@435 291 void VM_Version::print_features() {
duke@435 292 tty->print_cr("Version:%s", cpu_features());
duke@435 293 }
duke@435 294
duke@435 295 int VM_Version::determine_features() {
duke@435 296 if (UseV8InstrsOnly) {
duke@435 297 NOT_PRODUCT(if (PrintMiscellaneous && Verbose) tty->print_cr("Version is Forced-V8");)
duke@435 298 return generic_v8_m;
duke@435 299 }
duke@435 300
duke@435 301 int features = platform_features(unknown_m); // platform_features() is os_arch specific
duke@435 302
duke@435 303 if (features == unknown_m) {
duke@435 304 features = generic_v9_m;
duke@435 305 warning("Cannot recognize SPARC version. Default to V9");
duke@435 306 }
duke@435 307
kvn@2403 308 assert(is_T_family(features) == is_niagara(features), "Niagara should be T series");
kvn@2403 309 if (UseNiagaraInstrs) { // Force code generation for Niagara
kvn@2403 310 if (is_T_family(features)) {
duke@435 311 // Happy to accomodate...
duke@435 312 } else {
duke@435 313 NOT_PRODUCT(if (PrintMiscellaneous && Verbose) tty->print_cr("Version is Forced-Niagara");)
kvn@2403 314 features |= T_family_m;
duke@435 315 }
duke@435 316 } else {
kvn@2403 317 if (is_T_family(features) && !FLAG_IS_DEFAULT(UseNiagaraInstrs)) {
duke@435 318 NOT_PRODUCT(if (PrintMiscellaneous && Verbose) tty->print_cr("Version is Forced-Not-Niagara");)
kvn@2403 319 features &= ~(T_family_m | T1_model_m);
duke@435 320 } else {
duke@435 321 // Happy to accomodate...
duke@435 322 }
duke@435 323 }
duke@435 324
duke@435 325 return features;
duke@435 326 }
duke@435 327
duke@435 328 static int saved_features = 0;
duke@435 329
duke@435 330 void VM_Version::allow_all() {
duke@435 331 saved_features = _features;
duke@435 332 _features = all_features_m;
duke@435 333 }
duke@435 334
duke@435 335 void VM_Version::revert() {
duke@435 336 _features = saved_features;
duke@435 337 }
jmasa@445 338
jmasa@445 339 unsigned int VM_Version::calc_parallel_worker_threads() {
jmasa@445 340 unsigned int result;
twisti@4108 341 if (is_M_series()) {
twisti@4108 342 // for now, use same gc thread calculation for M-series as for niagara-plus
twisti@4108 343 // in future, we may want to tweak parameters for nof_parallel_worker_thread
twisti@4108 344 result = nof_parallel_worker_threads(5, 16, 8);
twisti@4108 345 } else if (is_niagara_plus()) {
jmasa@445 346 result = nof_parallel_worker_threads(5, 16, 8);
jmasa@445 347 } else {
jmasa@445 348 result = nof_parallel_worker_threads(5, 8, 8);
jmasa@445 349 }
jmasa@445 350 return result;
jmasa@445 351 }

mercurial