src/cpu/x86/vm/vm_version_x86_64.hpp

Sat, 01 Dec 2007 00:00:00 +0000

author
duke
date
Sat, 01 Dec 2007 00:00:00 +0000
changeset 435
a61af66fc99e
child 840
2649e5276dd7
permissions
-rw-r--r--

Initial load

duke@435 1 /*
duke@435 2 * Copyright 2003-2006 Sun Microsystems, Inc. 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 *
duke@435 19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
duke@435 20 * CA 95054 USA or visit www.sun.com if you need additional information or
duke@435 21 * have any questions.
duke@435 22 *
duke@435 23 */
duke@435 24
duke@435 25 class VM_Version : public Abstract_VM_Version {
duke@435 26 public:
duke@435 27 // cpuid result register layouts. These are all unions of a uint32_t
duke@435 28 // (in case anyone wants access to the register as a whole) and a bitfield.
duke@435 29
duke@435 30 union StdCpuid1Eax {
duke@435 31 uint32_t value;
duke@435 32 struct {
duke@435 33 uint32_t stepping : 4,
duke@435 34 model : 4,
duke@435 35 family : 4,
duke@435 36 proc_type : 2,
duke@435 37 : 2,
duke@435 38 ext_model : 4,
duke@435 39 ext_family : 8,
duke@435 40 : 4;
duke@435 41 } bits;
duke@435 42 };
duke@435 43
duke@435 44 union StdCpuid1Ebx { // example, unused
duke@435 45 uint32_t value;
duke@435 46 struct {
duke@435 47 uint32_t brand_id : 8,
duke@435 48 clflush_size : 8,
duke@435 49 threads_per_cpu : 8,
duke@435 50 apic_id : 8;
duke@435 51 } bits;
duke@435 52 };
duke@435 53
duke@435 54 union StdCpuid1Ecx {
duke@435 55 uint32_t value;
duke@435 56 struct {
duke@435 57 uint32_t sse3 : 1,
duke@435 58 : 2,
duke@435 59 monitor : 1,
duke@435 60 : 1,
duke@435 61 vmx : 1,
duke@435 62 : 1,
duke@435 63 est : 1,
duke@435 64 : 1,
duke@435 65 ssse3 : 1,
duke@435 66 cid : 1,
duke@435 67 : 2,
duke@435 68 cmpxchg16: 1,
duke@435 69 : 4,
duke@435 70 dca : 1,
duke@435 71 : 4,
duke@435 72 popcnt : 1,
duke@435 73 : 8;
duke@435 74 } bits;
duke@435 75 };
duke@435 76
duke@435 77 union StdCpuid1Edx {
duke@435 78 uint32_t value;
duke@435 79 struct {
duke@435 80 uint32_t : 4,
duke@435 81 tsc : 1,
duke@435 82 : 3,
duke@435 83 cmpxchg8 : 1,
duke@435 84 : 6,
duke@435 85 cmov : 1,
duke@435 86 : 7,
duke@435 87 mmx : 1,
duke@435 88 fxsr : 1,
duke@435 89 sse : 1,
duke@435 90 sse2 : 1,
duke@435 91 : 1,
duke@435 92 ht : 1,
duke@435 93 : 3;
duke@435 94 } bits;
duke@435 95 };
duke@435 96
duke@435 97 union DcpCpuid4Eax {
duke@435 98 uint32_t value;
duke@435 99 struct {
duke@435 100 uint32_t cache_type : 5,
duke@435 101 : 21,
duke@435 102 cores_per_cpu : 6;
duke@435 103 } bits;
duke@435 104 };
duke@435 105
duke@435 106 union DcpCpuid4Ebx {
duke@435 107 uint32_t value;
duke@435 108 struct {
duke@435 109 uint32_t L1_line_size : 12,
duke@435 110 partitions : 10,
duke@435 111 associativity : 10;
duke@435 112 } bits;
duke@435 113 };
duke@435 114
duke@435 115 union ExtCpuid1Edx {
duke@435 116 uint32_t value;
duke@435 117 struct {
duke@435 118 uint32_t : 22,
duke@435 119 mmx_amd : 1,
duke@435 120 mmx : 1,
duke@435 121 fxsr : 1,
duke@435 122 : 4,
duke@435 123 long_mode : 1,
duke@435 124 tdnow2 : 1,
duke@435 125 tdnow : 1;
duke@435 126 } bits;
duke@435 127 };
duke@435 128
duke@435 129 union ExtCpuid1Ecx {
duke@435 130 uint32_t value;
duke@435 131 struct {
duke@435 132 uint32_t LahfSahf : 1,
duke@435 133 CmpLegacy : 1,
duke@435 134 : 4,
duke@435 135 abm : 1,
duke@435 136 sse4a : 1,
duke@435 137 misalignsse : 1,
duke@435 138 prefetchw : 1,
duke@435 139 : 22;
duke@435 140 } bits;
duke@435 141 };
duke@435 142
duke@435 143 union ExtCpuid5Ex {
duke@435 144 uint32_t value;
duke@435 145 struct {
duke@435 146 uint32_t L1_line_size : 8,
duke@435 147 L1_tag_lines : 8,
duke@435 148 L1_assoc : 8,
duke@435 149 L1_size : 8;
duke@435 150 } bits;
duke@435 151 };
duke@435 152
duke@435 153 union ExtCpuid8Ecx {
duke@435 154 uint32_t value;
duke@435 155 struct {
duke@435 156 uint32_t cores_per_cpu : 8,
duke@435 157 : 24;
duke@435 158 } bits;
duke@435 159 };
duke@435 160
duke@435 161 protected:
duke@435 162 static int _cpu;
duke@435 163 static int _model;
duke@435 164 static int _stepping;
duke@435 165 static int _cpuFeatures; // features returned by the "cpuid" instruction
duke@435 166 // 0 if this instruction is not available
duke@435 167 static const char* _features_str;
duke@435 168
duke@435 169 enum {
duke@435 170 CPU_CX8 = (1 << 0), // next bits are from cpuid 1 (EDX)
duke@435 171 CPU_CMOV = (1 << 1),
duke@435 172 CPU_FXSR = (1 << 2),
duke@435 173 CPU_HT = (1 << 3),
duke@435 174 CPU_MMX = (1 << 4),
duke@435 175 CPU_3DNOW= (1 << 5),
duke@435 176 CPU_SSE = (1 << 6),
duke@435 177 CPU_SSE2 = (1 << 7),
duke@435 178 CPU_SSE3 = (1 << 8),
duke@435 179 CPU_SSSE3= (1 << 9),
duke@435 180 CPU_SSE4 = (1 <<10),
duke@435 181 CPU_SSE4A= (1 <<11)
duke@435 182 } cpuFeatureFlags;
duke@435 183
duke@435 184 // cpuid information block. All info derived from executing cpuid with
duke@435 185 // various function numbers is stored here. Intel and AMD info is
duke@435 186 // merged in this block: accessor methods disentangle it.
duke@435 187 //
duke@435 188 // The info block is laid out in subblocks of 4 dwords corresponding to
duke@435 189 // eax, ebx, ecx and edx, whether or not they contain anything useful.
duke@435 190 struct CpuidInfo {
duke@435 191 // cpuid function 0
duke@435 192 uint32_t std_max_function;
duke@435 193 uint32_t std_vendor_name_0;
duke@435 194 uint32_t std_vendor_name_1;
duke@435 195 uint32_t std_vendor_name_2;
duke@435 196
duke@435 197 // cpuid function 1
duke@435 198 StdCpuid1Eax std_cpuid1_eax;
duke@435 199 StdCpuid1Ebx std_cpuid1_ebx;
duke@435 200 StdCpuid1Ecx std_cpuid1_ecx;
duke@435 201 StdCpuid1Edx std_cpuid1_edx;
duke@435 202
duke@435 203 // cpuid function 4 (deterministic cache parameters)
duke@435 204 DcpCpuid4Eax dcp_cpuid4_eax;
duke@435 205 DcpCpuid4Ebx dcp_cpuid4_ebx;
duke@435 206 uint32_t dcp_cpuid4_ecx; // unused currently
duke@435 207 uint32_t dcp_cpuid4_edx; // unused currently
duke@435 208
duke@435 209 // cpuid function 0x80000000 // example, unused
duke@435 210 uint32_t ext_max_function;
duke@435 211 uint32_t ext_vendor_name_0;
duke@435 212 uint32_t ext_vendor_name_1;
duke@435 213 uint32_t ext_vendor_name_2;
duke@435 214
duke@435 215 // cpuid function 0x80000001
duke@435 216 uint32_t ext_cpuid1_eax; // reserved
duke@435 217 uint32_t ext_cpuid1_ebx; // reserved
duke@435 218 ExtCpuid1Ecx ext_cpuid1_ecx;
duke@435 219 ExtCpuid1Edx ext_cpuid1_edx;
duke@435 220
duke@435 221 // cpuid functions 0x80000002 thru 0x80000004: example, unused
duke@435 222 uint32_t proc_name_0, proc_name_1, proc_name_2, proc_name_3;
duke@435 223 uint32_t proc_name_4, proc_name_5, proc_name_6, proc_name_7;
duke@435 224 uint32_t proc_name_8, proc_name_9, proc_name_10,proc_name_11;
duke@435 225
duke@435 226 // cpuid function 0x80000005 //AMD L1, Intel reserved
duke@435 227 uint32_t ext_cpuid5_eax; // unused currently
duke@435 228 uint32_t ext_cpuid5_ebx; // reserved
duke@435 229 ExtCpuid5Ex ext_cpuid5_ecx; // L1 data cache info (AMD)
duke@435 230 ExtCpuid5Ex ext_cpuid5_edx; // L1 instruction cache info (AMD)
duke@435 231
duke@435 232 // cpuid function 0x80000008
duke@435 233 uint32_t ext_cpuid8_eax; // unused currently
duke@435 234 uint32_t ext_cpuid8_ebx; // reserved
duke@435 235 ExtCpuid8Ecx ext_cpuid8_ecx;
duke@435 236 uint32_t ext_cpuid8_edx; // reserved
duke@435 237 };
duke@435 238
duke@435 239 // The actual cpuid info block
duke@435 240 static CpuidInfo _cpuid_info;
duke@435 241
duke@435 242 // Extractors and predicates
duke@435 243 static bool is_extended_cpu_family() {
duke@435 244 const uint32_t Extended_Cpu_Family = 0xf;
duke@435 245 return _cpuid_info.std_cpuid1_eax.bits.family == Extended_Cpu_Family;
duke@435 246 }
duke@435 247 static uint32_t extended_cpu_family() {
duke@435 248 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.family;
duke@435 249 if (is_extended_cpu_family()) {
duke@435 250 result += _cpuid_info.std_cpuid1_eax.bits.ext_family;
duke@435 251 }
duke@435 252 return result;
duke@435 253 }
duke@435 254 static uint32_t extended_cpu_model() {
duke@435 255 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.model;
duke@435 256 if (is_extended_cpu_family()) {
duke@435 257 result |= _cpuid_info.std_cpuid1_eax.bits.ext_model << 4;
duke@435 258 }
duke@435 259 return result;
duke@435 260 }
duke@435 261 static uint32_t cpu_stepping() {
duke@435 262 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.stepping;
duke@435 263 return result;
duke@435 264 }
duke@435 265 static uint logical_processor_count() {
duke@435 266 uint result = threads_per_core();
duke@435 267 return result;
duke@435 268 }
duke@435 269 static uint32_t feature_flags() {
duke@435 270 uint32_t result = 0;
duke@435 271 if (_cpuid_info.std_cpuid1_edx.bits.cmpxchg8 != 0)
duke@435 272 result |= CPU_CX8;
duke@435 273 if (_cpuid_info.std_cpuid1_edx.bits.cmov != 0)
duke@435 274 result |= CPU_CMOV;
duke@435 275 if (_cpuid_info.std_cpuid1_edx.bits.fxsr != 0 || is_amd() &&
duke@435 276 _cpuid_info.ext_cpuid1_edx.bits.fxsr != 0)
duke@435 277 result |= CPU_FXSR;
duke@435 278 // HT flag is set for multi-core processors also.
duke@435 279 if (threads_per_core() > 1)
duke@435 280 result |= CPU_HT;
duke@435 281 if (_cpuid_info.std_cpuid1_edx.bits.mmx != 0 || is_amd() &&
duke@435 282 _cpuid_info.ext_cpuid1_edx.bits.mmx != 0)
duke@435 283 result |= CPU_MMX;
duke@435 284 if (is_amd() && _cpuid_info.ext_cpuid1_edx.bits.tdnow != 0)
duke@435 285 result |= CPU_3DNOW;
duke@435 286 if (_cpuid_info.std_cpuid1_edx.bits.sse != 0)
duke@435 287 result |= CPU_SSE;
duke@435 288 if (_cpuid_info.std_cpuid1_edx.bits.sse2 != 0)
duke@435 289 result |= CPU_SSE2;
duke@435 290 if (_cpuid_info.std_cpuid1_ecx.bits.sse3 != 0)
duke@435 291 result |= CPU_SSE3;
duke@435 292 if (_cpuid_info.std_cpuid1_ecx.bits.ssse3 != 0)
duke@435 293 result |= CPU_SSSE3;
duke@435 294 if (is_amd() && _cpuid_info.ext_cpuid1_ecx.bits.sse4a != 0)
duke@435 295 result |= CPU_SSE4A;
duke@435 296 return result;
duke@435 297 }
duke@435 298
duke@435 299 static void get_processor_features();
duke@435 300
duke@435 301 public:
duke@435 302 // Offsets for cpuid asm stub
duke@435 303 static ByteSize std_cpuid0_offset() { return byte_offset_of(CpuidInfo, std_max_function); }
duke@435 304 static ByteSize std_cpuid1_offset() { return byte_offset_of(CpuidInfo, std_cpuid1_eax); }
duke@435 305 static ByteSize dcp_cpuid4_offset() { return byte_offset_of(CpuidInfo, dcp_cpuid4_eax); }
duke@435 306 static ByteSize ext_cpuid1_offset() { return byte_offset_of(CpuidInfo, ext_cpuid1_eax); }
duke@435 307 static ByteSize ext_cpuid5_offset() { return byte_offset_of(CpuidInfo, ext_cpuid5_eax); }
duke@435 308 static ByteSize ext_cpuid8_offset() { return byte_offset_of(CpuidInfo, ext_cpuid8_eax); }
duke@435 309
duke@435 310 // Initialization
duke@435 311 static void initialize();
duke@435 312
duke@435 313 // Asserts
duke@435 314 static void assert_is_initialized() {
duke@435 315 assert(_cpuid_info.std_cpuid1_eax.bits.family != 0, "VM_Version not initialized");
duke@435 316 }
duke@435 317
duke@435 318 //
duke@435 319 // Processor family:
duke@435 320 // 3 - 386
duke@435 321 // 4 - 486
duke@435 322 // 5 - Pentium
duke@435 323 // 6 - PentiumPro, Pentium II, Celeron, Xeon, Pentium III, Athlon,
duke@435 324 // Pentium M, Core Solo, Core Duo, Core2 Duo
duke@435 325 // family 6 model: 9, 13, 14, 15
duke@435 326 // 0x0f - Pentium 4, Opteron
duke@435 327 //
duke@435 328 // Note: The cpu family should be used to select between
duke@435 329 // instruction sequences which are valid on all Intel
duke@435 330 // processors. Use the feature test functions below to
duke@435 331 // determine whether a particular instruction is supported.
duke@435 332 //
duke@435 333 static int cpu_family() { return _cpu;}
duke@435 334 static bool is_P6() { return cpu_family() >= 6; }
duke@435 335
duke@435 336 static bool is_amd() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x68747541; } // 'htuA'
duke@435 337 static bool is_intel() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x756e6547; } // 'uneG'
duke@435 338
duke@435 339 static uint cores_per_cpu() {
duke@435 340 uint result = 1;
duke@435 341 if (is_intel()) {
duke@435 342 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1);
duke@435 343 } else if (is_amd()) {
duke@435 344 result = (_cpuid_info.ext_cpuid8_ecx.bits.cores_per_cpu + 1);
duke@435 345 }
duke@435 346 return result;
duke@435 347 }
duke@435 348
duke@435 349 static uint threads_per_core() {
duke@435 350 uint result = 1;
duke@435 351 if (_cpuid_info.std_cpuid1_edx.bits.ht != 0) {
duke@435 352 result = _cpuid_info.std_cpuid1_ebx.bits.threads_per_cpu /
duke@435 353 cores_per_cpu();
duke@435 354 }
duke@435 355 return result;
duke@435 356 }
duke@435 357
duke@435 358 static intx L1_data_cache_line_size() {
duke@435 359 intx result = 0;
duke@435 360 if (is_intel()) {
duke@435 361 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1);
duke@435 362 } else if (is_amd()) {
duke@435 363 result = _cpuid_info.ext_cpuid5_ecx.bits.L1_line_size;
duke@435 364 }
duke@435 365 if (result < 32) // not defined ?
duke@435 366 result = 32; // 32 bytes by default for other x64
duke@435 367 return result;
duke@435 368 }
duke@435 369
duke@435 370 //
duke@435 371 // Feature identification
duke@435 372 //
duke@435 373 static bool supports_cpuid() { return _cpuFeatures != 0; }
duke@435 374 static bool supports_cmpxchg8() { return (_cpuFeatures & CPU_CX8) != 0; }
duke@435 375 static bool supports_cmov() { return (_cpuFeatures & CPU_CMOV) != 0; }
duke@435 376 static bool supports_fxsr() { return (_cpuFeatures & CPU_FXSR) != 0; }
duke@435 377 static bool supports_ht() { return (_cpuFeatures & CPU_HT) != 0; }
duke@435 378 static bool supports_mmx() { return (_cpuFeatures & CPU_MMX) != 0; }
duke@435 379 static bool supports_sse() { return (_cpuFeatures & CPU_SSE) != 0; }
duke@435 380 static bool supports_sse2() { return (_cpuFeatures & CPU_SSE2) != 0; }
duke@435 381 static bool supports_sse3() { return (_cpuFeatures & CPU_SSE3) != 0; }
duke@435 382 static bool supports_ssse3() { return (_cpuFeatures & CPU_SSSE3)!= 0; }
duke@435 383 static bool supports_sse4() { return (_cpuFeatures & CPU_SSE4) != 0; }
duke@435 384 //
duke@435 385 // AMD features
duke@435 386 //
duke@435 387 static bool supports_3dnow() { return (_cpuFeatures & CPU_3DNOW) != 0; }
duke@435 388 static bool supports_mmx_ext() { return is_amd() && _cpuid_info.ext_cpuid1_edx.bits.mmx_amd != 0; }
duke@435 389 static bool supports_3dnow2() { return is_amd() && _cpuid_info.ext_cpuid1_edx.bits.tdnow2 != 0; }
duke@435 390 static bool supports_sse4a() { return (_cpuFeatures & CPU_SSE4A) != 0; }
duke@435 391
duke@435 392 static bool supports_compare_and_exchange() { return true; }
duke@435 393
duke@435 394 static const char* cpu_features() { return _features_str; }
duke@435 395
duke@435 396 static intx allocate_prefetch_distance() {
duke@435 397 // This method should be called before allocate_prefetch_style().
duke@435 398 //
duke@435 399 // Hardware prefetching (distance/size in bytes):
duke@435 400 // Pentium 4 - 256 / 128
duke@435 401 // Opteron - 128 / 64 only when 2 sequential cache lines accessed
duke@435 402 // Core - 128 / 64
duke@435 403 //
duke@435 404 // Software prefetching (distance in bytes / instruction with best score):
duke@435 405 // Pentium 4 - 512 / prefetchnta
duke@435 406 // Opteron - 256 / prefetchnta
duke@435 407 // Core - 256 / prefetchnta
duke@435 408 // It will be used only when AllocatePrefetchStyle > 0
duke@435 409
duke@435 410 intx count = AllocatePrefetchDistance;
duke@435 411 if (count < 0) { // default ?
duke@435 412 if (is_amd()) { // AMD
duke@435 413 count = 256; // Opteron
duke@435 414 } else { // Intel
duke@435 415 if (cpu_family() == 6) {
duke@435 416 count = 256;// Pentium M, Core, Core2
duke@435 417 } else {
duke@435 418 count = 512;// Pentium 4
duke@435 419 }
duke@435 420 }
duke@435 421 }
duke@435 422 return count;
duke@435 423 }
duke@435 424 static intx allocate_prefetch_style() {
duke@435 425 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive");
duke@435 426 // Return 0 if AllocatePrefetchDistance was not defined.
duke@435 427 return AllocatePrefetchDistance > 0 ? AllocatePrefetchStyle : 0;
duke@435 428 }
duke@435 429
duke@435 430 // Prefetch interval for gc copy/scan == 9 dcache lines. Derived from
duke@435 431 // 50-warehouse specjbb runs on a 2-way 1.8ghz opteron using a 4gb heap.
duke@435 432 // Tested intervals from 128 to 2048 in increments of 64 == one cache line.
duke@435 433 // 256 bytes (4 dcache lines) was the nearest runner-up to 576.
duke@435 434
duke@435 435 // gc copy/scan is disabled if prefetchw isn't supported, because
duke@435 436 // Prefetch::write emits an inlined prefetchw on Linux.
duke@435 437 // Do not use the 3dnow prefetchw instruction. It isn't supported on em64t.
duke@435 438 // The used prefetcht0 instruction works for both amd64 and em64t.
duke@435 439 static intx prefetch_copy_interval_in_bytes() {
duke@435 440 intx interval = PrefetchCopyIntervalInBytes;
duke@435 441 return interval >= 0 ? interval : 576;
duke@435 442 }
duke@435 443 static intx prefetch_scan_interval_in_bytes() {
duke@435 444 intx interval = PrefetchScanIntervalInBytes;
duke@435 445 return interval >= 0 ? interval : 576;
duke@435 446 }
duke@435 447 static intx prefetch_fields_ahead() {
duke@435 448 intx count = PrefetchFieldsAhead;
duke@435 449 return count >= 0 ? count : 1;
duke@435 450 }
duke@435 451 };

mercurial