1 /* 2 * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #ifndef CPU_X86_VM_VM_VERSION_X86_HPP 26 #define CPU_X86_VM_VM_VERSION_X86_HPP 27 28 #include "runtime/globals_extension.hpp" 29 #include "runtime/vm_version.hpp" 30 31 class VM_Version : public Abstract_VM_Version { 32 public: 33 // cpuid result register layouts. These are all unions of a uint32_t 34 // (in case anyone wants access to the register as a whole) and a bitfield. 35 36 union StdCpuid1Eax { 37 uint32_t value; 38 struct { 39 uint32_t stepping : 4, 40 model : 4, 41 family : 4, 42 proc_type : 2, 43 : 2, 44 ext_model : 4, 45 ext_family : 8, 46 : 4; 47 } bits; 48 }; 49 50 union StdCpuid1Ebx { // example, unused 51 uint32_t value; 52 struct { 53 uint32_t brand_id : 8, 54 clflush_size : 8, 55 threads_per_cpu : 8, 56 apic_id : 8; 57 } bits; 58 }; 59 60 union StdCpuid1Ecx { 61 uint32_t value; 62 struct { 63 uint32_t sse3 : 1, 64 : 2, 65 monitor : 1, 66 : 1, 67 vmx : 1, 68 : 1, 69 est : 1, 70 : 1, 71 ssse3 : 1, 72 cid : 1, 73 : 2, 74 cmpxchg16: 1, 75 : 4, 76 dca : 1, 77 sse4_1 : 1, 78 sse4_2 : 1, 79 : 2, 80 popcnt : 1, 81 : 3, 82 osxsave : 1, 83 avx : 1, 84 : 3; 85 } bits; 86 }; 87 88 union StdCpuid1Edx { 89 uint32_t value; 90 struct { 91 uint32_t : 4, 92 tsc : 1, 93 : 3, 94 cmpxchg8 : 1, 95 : 6, 96 cmov : 1, 97 : 3, 98 clflush : 1, 99 : 3, 100 mmx : 1, 101 fxsr : 1, 102 sse : 1, 103 sse2 : 1, 104 : 1, 105 ht : 1, 106 : 3; 107 } bits; 108 }; 109 110 union DcpCpuid4Eax { 111 uint32_t value; 112 struct { 113 uint32_t cache_type : 5, 114 : 21, 115 cores_per_cpu : 6; 116 } bits; 117 }; 118 119 union DcpCpuid4Ebx { 120 uint32_t value; 121 struct { 122 uint32_t L1_line_size : 12, 123 partitions : 10, 124 associativity : 10; 125 } bits; 126 }; 127 128 union TplCpuidBEbx { 129 uint32_t value; 130 struct { 131 uint32_t logical_cpus : 16, 132 : 16; 133 } bits; 134 }; 135 136 union ExtCpuid1Ecx { 137 uint32_t value; 138 struct { 139 uint32_t LahfSahf : 1, 140 CmpLegacy : 1, 141 : 4, 142 lzcnt : 1, 143 sse4a : 1, 144 misalignsse : 1, 145 prefetchw : 1, 146 : 22; 147 } bits; 148 }; 149 150 union ExtCpuid1Edx { 151 uint32_t value; 152 struct { 153 uint32_t : 22, 154 mmx_amd : 1, 155 mmx : 1, 156 fxsr : 1, 157 : 4, 158 long_mode : 1, 159 tdnow2 : 1, 160 tdnow : 1; 161 } bits; 162 }; 163 164 union ExtCpuid5Ex { 165 uint32_t value; 166 struct { 167 uint32_t L1_line_size : 8, 168 L1_tag_lines : 8, 169 L1_assoc : 8, 170 L1_size : 8; 171 } bits; 172 }; 173 174 union ExtCpuid7Edx { 175 uint32_t value; 176 struct { 177 uint32_t : 8, 178 tsc_invariance : 1, 179 : 23; 180 } bits; 181 }; 182 183 union ExtCpuid8Ecx { 184 uint32_t value; 185 struct { 186 uint32_t cores_per_cpu : 8, 187 : 24; 188 } bits; 189 }; 190 191 union SefCpuid7Eax { 192 uint32_t value; 193 }; 194 195 union SefCpuid7Ebx { 196 uint32_t value; 197 struct { 198 uint32_t fsgsbase : 1, 199 : 2, 200 bmi1 : 1, 201 : 1, 202 avx2 : 1, 203 : 2, 204 bmi2 : 1, 205 : 23; 206 } bits; 207 }; 208 209 union XemXcr0Eax { 210 uint32_t value; 211 struct { 212 uint32_t x87 : 1, 213 sse : 1, 214 ymm : 1, 215 : 29; 216 } bits; 217 }; 218 219 protected: 220 static int _cpu; 221 static int _model; 222 static int _stepping; 223 static int _cpuFeatures; // features returned by the "cpuid" instruction 224 // 0 if this instruction is not available 225 static const char* _features_str; 226 227 enum { 228 CPU_CX8 = (1 << 0), // next bits are from cpuid 1 (EDX) 229 CPU_CMOV = (1 << 1), 230 CPU_FXSR = (1 << 2), 231 CPU_HT = (1 << 3), 232 CPU_MMX = (1 << 4), 233 CPU_3DNOW_PREFETCH = (1 << 5), // Processor supports 3dnow prefetch and prefetchw instructions 234 // may not necessarily support other 3dnow instructions 235 CPU_SSE = (1 << 6), 236 CPU_SSE2 = (1 << 7), 237 CPU_SSE3 = (1 << 8), // SSE3 comes from cpuid 1 (ECX) 238 CPU_SSSE3 = (1 << 9), 239 CPU_SSE4A = (1 << 10), 240 CPU_SSE4_1 = (1 << 11), 241 CPU_SSE4_2 = (1 << 12), 242 CPU_POPCNT = (1 << 13), 243 CPU_LZCNT = (1 << 14), 244 CPU_TSC = (1 << 15), 245 CPU_TSCINV = (1 << 16), 246 CPU_AVX = (1 << 17), 247 CPU_AVX2 = (1 << 18) 248 } cpuFeatureFlags; 249 250 enum { 251 // AMD 252 CPU_FAMILY_AMD_11H = 17, 253 // Intel 254 CPU_FAMILY_INTEL_CORE = 6, 255 CPU_MODEL_NEHALEM_EP = 26, 256 CPU_MODEL_WESTMERE_EP = 44, 257 // CPU_MODEL_IVYBRIDGE_EP = ??, TODO - get real value 258 CPU_MODEL_SANDYBRIDGE_EP = 45 259 } cpuExtendedFamily; 260 261 // cpuid information block. All info derived from executing cpuid with 262 // various function numbers is stored here. Intel and AMD info is 263 // merged in this block: accessor methods disentangle it. 264 // 265 // The info block is laid out in subblocks of 4 dwords corresponding to 266 // eax, ebx, ecx and edx, whether or not they contain anything useful. 267 struct CpuidInfo { 268 // cpuid function 0 269 uint32_t std_max_function; 270 uint32_t std_vendor_name_0; 271 uint32_t std_vendor_name_1; 272 uint32_t std_vendor_name_2; 273 274 // cpuid function 1 275 StdCpuid1Eax std_cpuid1_eax; 276 StdCpuid1Ebx std_cpuid1_ebx; 277 StdCpuid1Ecx std_cpuid1_ecx; 278 StdCpuid1Edx std_cpuid1_edx; 279 280 // cpuid function 4 (deterministic cache parameters) 281 DcpCpuid4Eax dcp_cpuid4_eax; 282 DcpCpuid4Ebx dcp_cpuid4_ebx; 283 uint32_t dcp_cpuid4_ecx; // unused currently 284 uint32_t dcp_cpuid4_edx; // unused currently 285 286 // cpuid function 7 (structured extended features) 287 SefCpuid7Eax sef_cpuid7_eax; 288 SefCpuid7Ebx sef_cpuid7_ebx; 289 uint32_t sef_cpuid7_ecx; // unused currently 290 uint32_t sef_cpuid7_edx; // unused currently 291 292 // cpuid function 0xB (processor topology) 293 // ecx = 0 294 uint32_t tpl_cpuidB0_eax; 295 TplCpuidBEbx tpl_cpuidB0_ebx; 296 uint32_t tpl_cpuidB0_ecx; // unused currently 297 uint32_t tpl_cpuidB0_edx; // unused currently 298 299 // ecx = 1 300 uint32_t tpl_cpuidB1_eax; 301 TplCpuidBEbx tpl_cpuidB1_ebx; 302 uint32_t tpl_cpuidB1_ecx; // unused currently 303 uint32_t tpl_cpuidB1_edx; // unused currently 304 305 // ecx = 2 306 uint32_t tpl_cpuidB2_eax; 307 TplCpuidBEbx tpl_cpuidB2_ebx; 308 uint32_t tpl_cpuidB2_ecx; // unused currently 309 uint32_t tpl_cpuidB2_edx; // unused currently 310 311 // cpuid function 0x80000000 // example, unused 312 uint32_t ext_max_function; 313 uint32_t ext_vendor_name_0; 314 uint32_t ext_vendor_name_1; 315 uint32_t ext_vendor_name_2; 316 317 // cpuid function 0x80000001 318 uint32_t ext_cpuid1_eax; // reserved 319 uint32_t ext_cpuid1_ebx; // reserved 320 ExtCpuid1Ecx ext_cpuid1_ecx; 321 ExtCpuid1Edx ext_cpuid1_edx; 322 323 // cpuid functions 0x80000002 thru 0x80000004: example, unused 324 uint32_t proc_name_0, proc_name_1, proc_name_2, proc_name_3; 325 uint32_t proc_name_4, proc_name_5, proc_name_6, proc_name_7; 326 uint32_t proc_name_8, proc_name_9, proc_name_10,proc_name_11; 327 328 // cpuid function 0x80000005 //AMD L1, Intel reserved 329 uint32_t ext_cpuid5_eax; // unused currently 330 uint32_t ext_cpuid5_ebx; // reserved 331 ExtCpuid5Ex ext_cpuid5_ecx; // L1 data cache info (AMD) 332 ExtCpuid5Ex ext_cpuid5_edx; // L1 instruction cache info (AMD) 333 334 // cpuid function 0x80000007 335 uint32_t ext_cpuid7_eax; // reserved 336 uint32_t ext_cpuid7_ebx; // reserved 337 uint32_t ext_cpuid7_ecx; // reserved 338 ExtCpuid7Edx ext_cpuid7_edx; // tscinv 339 340 // cpuid function 0x80000008 341 uint32_t ext_cpuid8_eax; // unused currently 342 uint32_t ext_cpuid8_ebx; // reserved 343 ExtCpuid8Ecx ext_cpuid8_ecx; 344 uint32_t ext_cpuid8_edx; // reserved 345 346 // extended control register XCR0 (the XFEATURE_ENABLED_MASK register) 347 XemXcr0Eax xem_xcr0_eax; 348 uint32_t xem_xcr0_edx; // reserved 349 }; 350 351 // The actual cpuid info block 352 static CpuidInfo _cpuid_info; 353 354 // Extractors and predicates 355 static uint32_t extended_cpu_family() { 356 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.family; 357 result += _cpuid_info.std_cpuid1_eax.bits.ext_family; 358 return result; 359 } 360 361 static uint32_t extended_cpu_model() { 362 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.model; 363 result |= _cpuid_info.std_cpuid1_eax.bits.ext_model << 4; 364 return result; 365 } 366 367 static uint32_t cpu_stepping() { 368 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.stepping; 369 return result; 370 } 371 372 static uint logical_processor_count() { 373 uint result = threads_per_core(); 374 return result; 375 } 376 377 static uint32_t feature_flags() { 378 uint32_t result = 0; 379 if (_cpuid_info.std_cpuid1_edx.bits.cmpxchg8 != 0) 380 result |= CPU_CX8; 381 if (_cpuid_info.std_cpuid1_edx.bits.cmov != 0) 382 result |= CPU_CMOV; 383 if (_cpuid_info.std_cpuid1_edx.bits.fxsr != 0 || (is_amd() && 384 _cpuid_info.ext_cpuid1_edx.bits.fxsr != 0)) 385 result |= CPU_FXSR; 386 // HT flag is set for multi-core processors also. 387 if (threads_per_core() > 1) 388 result |= CPU_HT; 389 if (_cpuid_info.std_cpuid1_edx.bits.mmx != 0 || (is_amd() && 390 _cpuid_info.ext_cpuid1_edx.bits.mmx != 0)) 391 result |= CPU_MMX; 392 if (_cpuid_info.std_cpuid1_edx.bits.sse != 0) 393 result |= CPU_SSE; 394 if (_cpuid_info.std_cpuid1_edx.bits.sse2 != 0) 395 result |= CPU_SSE2; 396 if (_cpuid_info.std_cpuid1_ecx.bits.sse3 != 0) 397 result |= CPU_SSE3; 398 if (_cpuid_info.std_cpuid1_ecx.bits.ssse3 != 0) 399 result |= CPU_SSSE3; 400 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_1 != 0) 401 result |= CPU_SSE4_1; 402 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_2 != 0) 403 result |= CPU_SSE4_2; 404 if (_cpuid_info.std_cpuid1_ecx.bits.popcnt != 0) 405 result |= CPU_POPCNT; 406 if (_cpuid_info.std_cpuid1_ecx.bits.avx != 0 && 407 _cpuid_info.std_cpuid1_ecx.bits.osxsave != 0 && 408 _cpuid_info.xem_xcr0_eax.bits.sse != 0 && 409 _cpuid_info.xem_xcr0_eax.bits.ymm != 0) { 410 result |= CPU_AVX; 411 if (_cpuid_info.sef_cpuid7_ebx.bits.avx2 != 0) 412 result |= CPU_AVX2; 413 } 414 if (_cpuid_info.std_cpuid1_edx.bits.tsc != 0) 415 result |= CPU_TSC; 416 if (_cpuid_info.ext_cpuid7_edx.bits.tsc_invariance != 0) 417 result |= CPU_TSCINV; 418 419 // AMD features. 420 if (is_amd()) { 421 if ((_cpuid_info.ext_cpuid1_edx.bits.tdnow != 0) || 422 (_cpuid_info.ext_cpuid1_ecx.bits.prefetchw != 0)) 423 result |= CPU_3DNOW_PREFETCH; 424 if (_cpuid_info.ext_cpuid1_ecx.bits.lzcnt != 0) 425 result |= CPU_LZCNT; 426 if (_cpuid_info.ext_cpuid1_ecx.bits.sse4a != 0) 427 result |= CPU_SSE4A; 428 } 429 430 return result; 431 } 432 433 static void get_processor_features(); 434 435 public: 436 // Offsets for cpuid asm stub 437 static ByteSize std_cpuid0_offset() { return byte_offset_of(CpuidInfo, std_max_function); } 438 static ByteSize std_cpuid1_offset() { return byte_offset_of(CpuidInfo, std_cpuid1_eax); } 439 static ByteSize dcp_cpuid4_offset() { return byte_offset_of(CpuidInfo, dcp_cpuid4_eax); } 440 static ByteSize sef_cpuid7_offset() { return byte_offset_of(CpuidInfo, sef_cpuid7_eax); } 441 static ByteSize ext_cpuid1_offset() { return byte_offset_of(CpuidInfo, ext_cpuid1_eax); } 442 static ByteSize ext_cpuid5_offset() { return byte_offset_of(CpuidInfo, ext_cpuid5_eax); } 443 static ByteSize ext_cpuid7_offset() { return byte_offset_of(CpuidInfo, ext_cpuid7_eax); } 444 static ByteSize ext_cpuid8_offset() { return byte_offset_of(CpuidInfo, ext_cpuid8_eax); } 445 static ByteSize tpl_cpuidB0_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB0_eax); } 446 static ByteSize tpl_cpuidB1_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB1_eax); } 447 static ByteSize tpl_cpuidB2_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB2_eax); } 448 static ByteSize xem_xcr0_offset() { return byte_offset_of(CpuidInfo, xem_xcr0_eax); } 449 450 // Initialization 451 static void initialize(); 452 453 // Asserts 454 static void assert_is_initialized() { 455 assert(_cpuid_info.std_cpuid1_eax.bits.family != 0, "VM_Version not initialized"); 456 } 457 458 // 459 // Processor family: 460 // 3 - 386 461 // 4 - 486 462 // 5 - Pentium 463 // 6 - PentiumPro, Pentium II, Celeron, Xeon, Pentium III, Athlon, 464 // Pentium M, Core Solo, Core Duo, Core2 Duo 465 // family 6 model: 9, 13, 14, 15 466 // 0x0f - Pentium 4, Opteron 467 // 468 // Note: The cpu family should be used to select between 469 // instruction sequences which are valid on all Intel 470 // processors. Use the feature test functions below to 471 // determine whether a particular instruction is supported. 472 // 473 static int cpu_family() { return _cpu;} 474 static bool is_P6() { return cpu_family() >= 6; } 475 static bool is_amd() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x68747541; } // 'htuA' 476 static bool is_intel() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x756e6547; } // 'uneG' 477 478 static bool supports_processor_topology() { 479 return (_cpuid_info.std_max_function >= 0xB) && 480 // eax[4:0] | ebx[0:15] == 0 indicates invalid topology level. 481 // Some cpus have max cpuid >= 0xB but do not support processor topology. 482 ((_cpuid_info.tpl_cpuidB0_eax & 0x1f | _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus) != 0); 483 } 484 485 static uint cores_per_cpu() { 486 uint result = 1; 487 if (is_intel()) { 488 if (supports_processor_topology()) { 489 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus / 490 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 491 } else { 492 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1); 493 } 494 } else if (is_amd()) { 495 result = (_cpuid_info.ext_cpuid8_ecx.bits.cores_per_cpu + 1); 496 } 497 return result; 498 } 499 500 static uint threads_per_core() { 501 uint result = 1; 502 if (is_intel() && supports_processor_topology()) { 503 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 504 } else if (_cpuid_info.std_cpuid1_edx.bits.ht != 0) { 505 result = _cpuid_info.std_cpuid1_ebx.bits.threads_per_cpu / 506 cores_per_cpu(); 507 } 508 return result; 509 } 510 511 static intx prefetch_data_size() { 512 intx result = 0; 513 if (is_intel()) { 514 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1); 515 } else if (is_amd()) { 516 result = _cpuid_info.ext_cpuid5_ecx.bits.L1_line_size; 517 } 518 if (result < 32) // not defined ? 519 result = 32; // 32 bytes by default on x86 and other x64 520 return result; 521 } 522 523 // 524 // Feature identification 525 // 526 static bool supports_cpuid() { return _cpuFeatures != 0; } 527 static bool supports_cmpxchg8() { return (_cpuFeatures & CPU_CX8) != 0; } 528 static bool supports_cmov() { return (_cpuFeatures & CPU_CMOV) != 0; } 529 static bool supports_fxsr() { return (_cpuFeatures & CPU_FXSR) != 0; } 530 static bool supports_ht() { return (_cpuFeatures & CPU_HT) != 0; } 531 static bool supports_mmx() { return (_cpuFeatures & CPU_MMX) != 0; } 532 static bool supports_sse() { return (_cpuFeatures & CPU_SSE) != 0; } 533 static bool supports_sse2() { return (_cpuFeatures & CPU_SSE2) != 0; } 534 static bool supports_sse3() { return (_cpuFeatures & CPU_SSE3) != 0; } 535 static bool supports_ssse3() { return (_cpuFeatures & CPU_SSSE3)!= 0; } 536 static bool supports_sse4_1() { return (_cpuFeatures & CPU_SSE4_1) != 0; } 537 static bool supports_sse4_2() { return (_cpuFeatures & CPU_SSE4_2) != 0; } 538 static bool supports_popcnt() { return (_cpuFeatures & CPU_POPCNT) != 0; } 539 static bool supports_avx() { return (_cpuFeatures & CPU_AVX) != 0; } 540 static bool supports_avx2() { return (_cpuFeatures & CPU_AVX2) != 0; } 541 static bool supports_tsc() { return (_cpuFeatures & CPU_TSC) != 0; } 542 543 // Intel features 544 static bool is_intel_family_core() { return is_intel() && 545 extended_cpu_family() == CPU_FAMILY_INTEL_CORE; } 546 547 static bool is_intel_tsc_synched_at_init() { 548 if (is_intel_family_core()) { 549 uint32_t ext_model = extended_cpu_model(); 550 if (ext_model == CPU_MODEL_NEHALEM_EP || 551 ext_model == CPU_MODEL_WESTMERE_EP || 552 // TODO ext_model == CPU_MODEL_IVYBRIDGE_EP || 553 ext_model == CPU_MODEL_SANDYBRIDGE_EP) { 554 // 2-socket invtsc support. EX versions with 4 sockets are not 555 // guaranteed to synchronize tscs at initialization via a double 556 // handshake. The tscs can be explicitly set in software. Code 557 // that uses tsc values must be prepared for them to arbitrarily 558 // jump backward or forward. 559 return true; 560 } 561 } 562 return false; 563 } 564 565 // AMD features 566 static bool supports_3dnow_prefetch() { return (_cpuFeatures & CPU_3DNOW_PREFETCH) != 0; } 567 static bool supports_mmx_ext() { return is_amd() && _cpuid_info.ext_cpuid1_edx.bits.mmx_amd != 0; } 568 static bool supports_lzcnt() { return (_cpuFeatures & CPU_LZCNT) != 0; } 569 static bool supports_sse4a() { return (_cpuFeatures & CPU_SSE4A) != 0; } 570 571 static bool is_amd_Barcelona() { return is_amd() && 572 extended_cpu_family() == CPU_FAMILY_AMD_11H; } 573 574 // Intel and AMD newer cores support fast timestamps well 575 static bool supports_tscinv_bit() { 576 return (_cpuFeatures & CPU_TSCINV) != 0; 577 } 578 static bool supports_tscinv() { 579 return supports_tscinv_bit() && 580 ( (is_amd() && !is_amd_Barcelona()) || 581 is_intel_tsc_synched_at_init() ); 582 } 583 584 // Intel Core and newer cpus have fast IDIV instruction (excluding Atom). 585 static bool has_fast_idiv() { return is_intel() && cpu_family() == 6 && 586 supports_sse3() && _model != 0x1C; } 587 588 static bool supports_compare_and_exchange() { return true; } 589 590 static const char* cpu_features() { return _features_str; } 591 592 static intx allocate_prefetch_distance() { 593 // This method should be called before allocate_prefetch_style(). 594 // 595 // Hardware prefetching (distance/size in bytes): 596 // Pentium 3 - 64 / 32 597 // Pentium 4 - 256 / 128 598 // Athlon - 64 / 32 ???? 599 // Opteron - 128 / 64 only when 2 sequential cache lines accessed 600 // Core - 128 / 64 601 // 602 // Software prefetching (distance in bytes / instruction with best score): 603 // Pentium 3 - 128 / prefetchnta 604 // Pentium 4 - 512 / prefetchnta 605 // Athlon - 128 / prefetchnta 606 // Opteron - 256 / prefetchnta 607 // Core - 256 / prefetchnta 608 // It will be used only when AllocatePrefetchStyle > 0 609 610 intx count = AllocatePrefetchDistance; 611 if (count < 0) { // default ? 612 if (is_amd()) { // AMD 613 if (supports_sse2()) 614 count = 256; // Opteron 615 else 616 count = 128; // Athlon 617 } else { // Intel 618 if (supports_sse2()) 619 if (cpu_family() == 6) { 620 count = 256; // Pentium M, Core, Core2 621 } else { 622 count = 512; // Pentium 4 623 } 624 else 625 count = 128; // Pentium 3 (and all other old CPUs) 626 } 627 } 628 return count; 629 } 630 static intx allocate_prefetch_style() { 631 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive"); 632 // Return 0 if AllocatePrefetchDistance was not defined. 633 return AllocatePrefetchDistance > 0 ? AllocatePrefetchStyle : 0; 634 } 635 636 // Prefetch interval for gc copy/scan == 9 dcache lines. Derived from 637 // 50-warehouse specjbb runs on a 2-way 1.8ghz opteron using a 4gb heap. 638 // Tested intervals from 128 to 2048 in increments of 64 == one cache line. 639 // 256 bytes (4 dcache lines) was the nearest runner-up to 576. 640 641 // gc copy/scan is disabled if prefetchw isn't supported, because 642 // Prefetch::write emits an inlined prefetchw on Linux. 643 // Do not use the 3dnow prefetchw instruction. It isn't supported on em64t. 644 // The used prefetcht0 instruction works for both amd64 and em64t. 645 static intx prefetch_copy_interval_in_bytes() { 646 intx interval = PrefetchCopyIntervalInBytes; 647 return interval >= 0 ? interval : 576; 648 } 649 static intx prefetch_scan_interval_in_bytes() { 650 intx interval = PrefetchScanIntervalInBytes; 651 return interval >= 0 ? interval : 576; 652 } 653 static intx prefetch_fields_ahead() { 654 intx count = PrefetchFieldsAhead; 655 return count >= 0 ? count : 1; 656 } 657 }; 658 659 #endif // CPU_X86_VM_VM_VERSION_X86_HPP