1 /* 2 * Copyright (c) 1997, 2012, 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 #include "precompiled.hpp" 26 #include "assembler_x86.inline.hpp" 27 #include "memory/resourceArea.hpp" 28 #include "runtime/java.hpp" 29 #include "runtime/stubCodeGenerator.hpp" 30 #include "vm_version_x86.hpp" 31 #ifdef TARGET_OS_FAMILY_linux 32 # include "os_linux.inline.hpp" 33 #endif 34 #ifdef TARGET_OS_FAMILY_solaris 35 # include "os_solaris.inline.hpp" 36 #endif 37 #ifdef TARGET_OS_FAMILY_windows 38 # include "os_windows.inline.hpp" 39 #endif 40 #ifdef TARGET_OS_FAMILY_bsd 41 # include "os_bsd.inline.hpp" 42 #endif 43 44 45 int VM_Version::_cpu; 46 int VM_Version::_model; 47 int VM_Version::_stepping; 48 int VM_Version::_cpuFeatures; 49 const char* VM_Version::_features_str = ""; 50 VM_Version::CpuidInfo VM_Version::_cpuid_info = { 0, }; 51 52 static BufferBlob* stub_blob; 53 static const int stub_size = 550; 54 55 extern "C" { 56 typedef void (*getPsrInfo_stub_t)(void*); 57 } 58 static getPsrInfo_stub_t getPsrInfo_stub = NULL; 59 60 61 class VM_Version_StubGenerator: public StubCodeGenerator { 62 public: 63 64 VM_Version_StubGenerator(CodeBuffer *c) : StubCodeGenerator(c) {} 65 66 address generate_getPsrInfo() { 67 // Flags to test CPU type. 68 const uint32_t HS_EFL_AC = 0x40000; 69 const uint32_t HS_EFL_ID = 0x200000; 70 // Values for when we don't have a CPUID instruction. 71 const int CPU_FAMILY_SHIFT = 8; 72 const uint32_t CPU_FAMILY_386 = (3 << CPU_FAMILY_SHIFT); 73 const uint32_t CPU_FAMILY_486 = (4 << CPU_FAMILY_SHIFT); 74 75 Label detect_486, cpu486, detect_586, std_cpuid1, std_cpuid4; 76 Label sef_cpuid, ext_cpuid, ext_cpuid1, ext_cpuid5, ext_cpuid7, done; 77 78 StubCodeMark mark(this, "VM_Version", "getPsrInfo_stub"); 79 # define __ _masm-> 80 81 address start = __ pc(); 82 83 // 84 // void getPsrInfo(VM_Version::CpuidInfo* cpuid_info); 85 // 86 // LP64: rcx and rdx are first and second argument registers on windows 87 88 __ push(rbp); 89 #ifdef _LP64 90 __ mov(rbp, c_rarg0); // cpuid_info address 91 #else 92 __ movptr(rbp, Address(rsp, 8)); // cpuid_info address 93 #endif 94 __ push(rbx); 95 __ push(rsi); 96 __ pushf(); // preserve rbx, and flags 97 __ pop(rax); 98 __ push(rax); 99 __ mov(rcx, rax); 100 // 101 // if we are unable to change the AC flag, we have a 386 102 // 103 __ xorl(rax, HS_EFL_AC); 104 __ push(rax); 105 __ popf(); 106 __ pushf(); 107 __ pop(rax); 108 __ cmpptr(rax, rcx); 109 __ jccb(Assembler::notEqual, detect_486); 110 111 __ movl(rax, CPU_FAMILY_386); 112 __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax); 113 __ jmp(done); 114 115 // 116 // If we are unable to change the ID flag, we have a 486 which does 117 // not support the "cpuid" instruction. 118 // 119 __ bind(detect_486); 120 __ mov(rax, rcx); 121 __ xorl(rax, HS_EFL_ID); 122 __ push(rax); 123 __ popf(); 124 __ pushf(); 125 __ pop(rax); 126 __ cmpptr(rcx, rax); 127 __ jccb(Assembler::notEqual, detect_586); 128 129 __ bind(cpu486); 130 __ movl(rax, CPU_FAMILY_486); 131 __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax); 132 __ jmp(done); 133 134 // 135 // At this point, we have a chip which supports the "cpuid" instruction 136 // 137 __ bind(detect_586); 138 __ xorl(rax, rax); 139 __ cpuid(); 140 __ orl(rax, rax); 141 __ jcc(Assembler::equal, cpu486); // if cpuid doesn't support an input 142 // value of at least 1, we give up and 143 // assume a 486 144 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); 145 __ movl(Address(rsi, 0), rax); 146 __ movl(Address(rsi, 4), rbx); 147 __ movl(Address(rsi, 8), rcx); 148 __ movl(Address(rsi,12), rdx); 149 150 __ cmpl(rax, 0xa); // Is cpuid(0xB) supported? 151 __ jccb(Assembler::belowEqual, std_cpuid4); 152 153 // 154 // cpuid(0xB) Processor Topology 155 // 156 __ movl(rax, 0xb); 157 __ xorl(rcx, rcx); // Threads level 158 __ cpuid(); 159 160 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB0_offset()))); 161 __ movl(Address(rsi, 0), rax); 162 __ movl(Address(rsi, 4), rbx); 163 __ movl(Address(rsi, 8), rcx); 164 __ movl(Address(rsi,12), rdx); 165 166 __ movl(rax, 0xb); 167 __ movl(rcx, 1); // Cores level 168 __ cpuid(); 169 __ push(rax); 170 __ andl(rax, 0x1f); // Determine if valid topology level 171 __ orl(rax, rbx); // eax[4:0] | ebx[0:15] == 0 indicates invalid level 172 __ andl(rax, 0xffff); 173 __ pop(rax); 174 __ jccb(Assembler::equal, std_cpuid4); 175 176 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB1_offset()))); 177 __ movl(Address(rsi, 0), rax); 178 __ movl(Address(rsi, 4), rbx); 179 __ movl(Address(rsi, 8), rcx); 180 __ movl(Address(rsi,12), rdx); 181 182 __ movl(rax, 0xb); 183 __ movl(rcx, 2); // Packages level 184 __ cpuid(); 185 __ push(rax); 186 __ andl(rax, 0x1f); // Determine if valid topology level 187 __ orl(rax, rbx); // eax[4:0] | ebx[0:15] == 0 indicates invalid level 188 __ andl(rax, 0xffff); 189 __ pop(rax); 190 __ jccb(Assembler::equal, std_cpuid4); 191 192 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB2_offset()))); 193 __ movl(Address(rsi, 0), rax); 194 __ movl(Address(rsi, 4), rbx); 195 __ movl(Address(rsi, 8), rcx); 196 __ movl(Address(rsi,12), rdx); 197 198 // 199 // cpuid(0x4) Deterministic cache params 200 // 201 __ bind(std_cpuid4); 202 __ movl(rax, 4); 203 __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x4) supported? 204 __ jccb(Assembler::greater, std_cpuid1); 205 206 __ xorl(rcx, rcx); // L1 cache 207 __ cpuid(); 208 __ push(rax); 209 __ andl(rax, 0x1f); // Determine if valid cache parameters used 210 __ orl(rax, rax); // eax[4:0] == 0 indicates invalid cache 211 __ pop(rax); 212 __ jccb(Assembler::equal, std_cpuid1); 213 214 __ lea(rsi, Address(rbp, in_bytes(VM_Version::dcp_cpuid4_offset()))); 215 __ movl(Address(rsi, 0), rax); 216 __ movl(Address(rsi, 4), rbx); 217 __ movl(Address(rsi, 8), rcx); 218 __ movl(Address(rsi,12), rdx); 219 220 // 221 // Standard cpuid(0x1) 222 // 223 __ bind(std_cpuid1); 224 __ movl(rax, 1); 225 __ cpuid(); 226 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset()))); 227 __ movl(Address(rsi, 0), rax); 228 __ movl(Address(rsi, 4), rbx); 229 __ movl(Address(rsi, 8), rcx); 230 __ movl(Address(rsi,12), rdx); 231 232 // 233 // Check if OS has enabled XGETBV instruction to access XCR0 234 // (OSXSAVE feature flag) and CPU supports AVX 235 // 236 __ andl(rcx, 0x18000000); 237 __ cmpl(rcx, 0x18000000); 238 __ jccb(Assembler::notEqual, sef_cpuid); 239 240 // 241 // XCR0, XFEATURE_ENABLED_MASK register 242 // 243 __ xorl(rcx, rcx); // zero for XCR0 register 244 __ xgetbv(); 245 __ lea(rsi, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); 246 __ movl(Address(rsi, 0), rax); 247 __ movl(Address(rsi, 4), rdx); 248 249 // 250 // cpuid(0x7) Structured Extended Features 251 // 252 __ bind(sef_cpuid); 253 __ movl(rax, 7); 254 __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x7) supported? 255 __ jccb(Assembler::greater, ext_cpuid); 256 257 __ xorl(rcx, rcx); 258 __ cpuid(); 259 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset()))); 260 __ movl(Address(rsi, 0), rax); 261 __ movl(Address(rsi, 4), rbx); 262 263 // 264 // Extended cpuid(0x80000000) 265 // 266 __ bind(ext_cpuid); 267 __ movl(rax, 0x80000000); 268 __ cpuid(); 269 __ cmpl(rax, 0x80000000); // Is cpuid(0x80000001) supported? 270 __ jcc(Assembler::belowEqual, done); 271 __ cmpl(rax, 0x80000004); // Is cpuid(0x80000005) supported? 272 __ jccb(Assembler::belowEqual, ext_cpuid1); 273 __ cmpl(rax, 0x80000006); // Is cpuid(0x80000007) supported? 274 __ jccb(Assembler::belowEqual, ext_cpuid5); 275 __ cmpl(rax, 0x80000007); // Is cpuid(0x80000008) supported? 276 __ jccb(Assembler::belowEqual, ext_cpuid7); 277 // 278 // Extended cpuid(0x80000008) 279 // 280 __ movl(rax, 0x80000008); 281 __ cpuid(); 282 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid8_offset()))); 283 __ movl(Address(rsi, 0), rax); 284 __ movl(Address(rsi, 4), rbx); 285 __ movl(Address(rsi, 8), rcx); 286 __ movl(Address(rsi,12), rdx); 287 288 // 289 // Extended cpuid(0x80000007) 290 // 291 __ bind(ext_cpuid7); 292 __ movl(rax, 0x80000007); 293 __ cpuid(); 294 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid7_offset()))); 295 __ movl(Address(rsi, 0), rax); 296 __ movl(Address(rsi, 4), rbx); 297 __ movl(Address(rsi, 8), rcx); 298 __ movl(Address(rsi,12), rdx); 299 300 // 301 // Extended cpuid(0x80000005) 302 // 303 __ bind(ext_cpuid5); 304 __ movl(rax, 0x80000005); 305 __ cpuid(); 306 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid5_offset()))); 307 __ movl(Address(rsi, 0), rax); 308 __ movl(Address(rsi, 4), rbx); 309 __ movl(Address(rsi, 8), rcx); 310 __ movl(Address(rsi,12), rdx); 311 312 // 313 // Extended cpuid(0x80000001) 314 // 315 __ bind(ext_cpuid1); 316 __ movl(rax, 0x80000001); 317 __ cpuid(); 318 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid1_offset()))); 319 __ movl(Address(rsi, 0), rax); 320 __ movl(Address(rsi, 4), rbx); 321 __ movl(Address(rsi, 8), rcx); 322 __ movl(Address(rsi,12), rdx); 323 324 // 325 // return 326 // 327 __ bind(done); 328 __ popf(); 329 __ pop(rsi); 330 __ pop(rbx); 331 __ pop(rbp); 332 __ ret(0); 333 334 # undef __ 335 336 return start; 337 }; 338 }; 339 340 341 void VM_Version::get_processor_features() { 342 343 _cpu = 4; // 486 by default 344 _model = 0; 345 _stepping = 0; 346 _cpuFeatures = 0; 347 _logical_processors_per_package = 1; 348 349 if (!Use486InstrsOnly) { 350 // Get raw processor info 351 getPsrInfo_stub(&_cpuid_info); 352 assert_is_initialized(); 353 _cpu = extended_cpu_family(); 354 _model = extended_cpu_model(); 355 _stepping = cpu_stepping(); 356 357 if (cpu_family() > 4) { // it supports CPUID 358 _cpuFeatures = feature_flags(); 359 // Logical processors are only available on P4s and above, 360 // and only if hyperthreading is available. 361 _logical_processors_per_package = logical_processor_count(); 362 } 363 } 364 365 _supports_cx8 = supports_cmpxchg8(); 366 367 #ifdef _LP64 368 // OS should support SSE for x64 and hardware should support at least SSE2. 369 if (!VM_Version::supports_sse2()) { 370 vm_exit_during_initialization("Unknown x64 processor: SSE2 not supported"); 371 } 372 // in 64 bit the use of SSE2 is the minimum 373 if (UseSSE < 2) UseSSE = 2; 374 #endif 375 376 #ifdef AMD64 377 // flush_icache_stub have to be generated first. 378 // That is why Icache line size is hard coded in ICache class, 379 // see icache_x86.hpp. It is also the reason why we can't use 380 // clflush instruction in 32-bit VM since it could be running 381 // on CPU which does not support it. 382 // 383 // The only thing we can do is to verify that flushed 384 // ICache::line_size has correct value. 385 guarantee(_cpuid_info.std_cpuid1_edx.bits.clflush != 0, "clflush is not supported"); 386 // clflush_size is size in quadwords (8 bytes). 387 guarantee(_cpuid_info.std_cpuid1_ebx.bits.clflush_size == 8, "such clflush size is not supported"); 388 #endif 389 390 // If the OS doesn't support SSE, we can't use this feature even if the HW does 391 if (!os::supports_sse()) 392 _cpuFeatures &= ~(CPU_SSE|CPU_SSE2|CPU_SSE3|CPU_SSSE3|CPU_SSE4A|CPU_SSE4_1|CPU_SSE4_2); 393 394 if (UseSSE < 4) { 395 _cpuFeatures &= ~CPU_SSE4_1; 396 _cpuFeatures &= ~CPU_SSE4_2; 397 } 398 399 if (UseSSE < 3) { 400 _cpuFeatures &= ~CPU_SSE3; 401 _cpuFeatures &= ~CPU_SSSE3; 402 _cpuFeatures &= ~CPU_SSE4A; 403 } 404 405 if (UseSSE < 2) 406 _cpuFeatures &= ~CPU_SSE2; 407 408 if (UseSSE < 1) 409 _cpuFeatures &= ~CPU_SSE; 410 411 if (UseAVX < 2) 412 _cpuFeatures &= ~CPU_AVX2; 413 414 if (UseAVX < 1) 415 _cpuFeatures &= ~CPU_AVX; 416 417 if (logical_processors_per_package() == 1) { 418 // HT processor could be installed on a system which doesn't support HT. 419 _cpuFeatures &= ~CPU_HT; 420 } 421 422 char buf[256]; 423 jio_snprintf(buf, sizeof(buf), "(%u cores per cpu, %u threads per core) family %d model %d stepping %d%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s", 424 cores_per_cpu(), threads_per_core(), 425 cpu_family(), _model, _stepping, 426 (supports_cmov() ? ", cmov" : ""), 427 (supports_cmpxchg8() ? ", cx8" : ""), 428 (supports_fxsr() ? ", fxsr" : ""), 429 (supports_mmx() ? ", mmx" : ""), 430 (supports_sse() ? ", sse" : ""), 431 (supports_sse2() ? ", sse2" : ""), 432 (supports_sse3() ? ", sse3" : ""), 433 (supports_ssse3()? ", ssse3": ""), 434 (supports_sse4_1() ? ", sse4.1" : ""), 435 (supports_sse4_2() ? ", sse4.2" : ""), 436 (supports_popcnt() ? ", popcnt" : ""), 437 (supports_avx() ? ", avx" : ""), 438 (supports_avx2() ? ", avx2" : ""), 439 (supports_mmx_ext() ? ", mmxext" : ""), 440 (supports_3dnow_prefetch() ? ", 3dnowpref" : ""), 441 (supports_lzcnt() ? ", lzcnt": ""), 442 (supports_sse4a() ? ", sse4a": ""), 443 (supports_ht() ? ", ht": ""), 444 (supports_tsc() ? ", tsc": ""), 445 (supports_tscinv_bit() ? ", tscinvbit": ""), 446 (supports_tscinv() ? ", tscinv": "")); 447 _features_str = strdup(buf); 448 449 // UseSSE is set to the smaller of what hardware supports and what 450 // the command line requires. I.e., you cannot set UseSSE to 2 on 451 // older Pentiums which do not support it. 452 if (UseSSE > 4) UseSSE=4; 453 if (UseSSE < 0) UseSSE=0; 454 if (!supports_sse4_1()) // Drop to 3 if no SSE4 support 455 UseSSE = MIN2((intx)3,UseSSE); 456 if (!supports_sse3()) // Drop to 2 if no SSE3 support 457 UseSSE = MIN2((intx)2,UseSSE); 458 if (!supports_sse2()) // Drop to 1 if no SSE2 support 459 UseSSE = MIN2((intx)1,UseSSE); 460 if (!supports_sse ()) // Drop to 0 if no SSE support 461 UseSSE = 0; 462 463 if (UseAVX > 2) UseAVX=2; 464 if (UseAVX < 0) UseAVX=0; 465 if (!supports_avx2()) // Drop to 1 if no AVX2 support 466 UseAVX = MIN2((intx)1,UseAVX); 467 if (!supports_avx ()) // Drop to 0 if no AVX support 468 UseAVX = 0; 469 470 // On new cpus instructions which update whole XMM register should be used 471 // to prevent partial register stall due to dependencies on high half. 472 // 473 // UseXmmLoadAndClearUpper == true --> movsd(xmm, mem) 474 // UseXmmLoadAndClearUpper == false --> movlpd(xmm, mem) 475 // UseXmmRegToRegMoveAll == true --> movaps(xmm, xmm), movapd(xmm, xmm). 476 // UseXmmRegToRegMoveAll == false --> movss(xmm, xmm), movsd(xmm, xmm). 477 478 if( is_amd() ) { // AMD cpus specific settings 479 if( supports_sse2() && FLAG_IS_DEFAULT(UseAddressNop) ) { 480 // Use it on new AMD cpus starting from Opteron. 481 UseAddressNop = true; 482 } 483 if( supports_sse2() && FLAG_IS_DEFAULT(UseNewLongLShift) ) { 484 // Use it on new AMD cpus starting from Opteron. 485 UseNewLongLShift = true; 486 } 487 if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) { 488 if( supports_sse4a() ) { 489 UseXmmLoadAndClearUpper = true; // use movsd only on '10h' Opteron 490 } else { 491 UseXmmLoadAndClearUpper = false; 492 } 493 } 494 if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) { 495 if( supports_sse4a() ) { 496 UseXmmRegToRegMoveAll = true; // use movaps, movapd only on '10h' 497 } else { 498 UseXmmRegToRegMoveAll = false; 499 } 500 } 501 if( FLAG_IS_DEFAULT(UseXmmI2F) ) { 502 if( supports_sse4a() ) { 503 UseXmmI2F = true; 504 } else { 505 UseXmmI2F = false; 506 } 507 } 508 if( FLAG_IS_DEFAULT(UseXmmI2D) ) { 509 if( supports_sse4a() ) { 510 UseXmmI2D = true; 511 } else { 512 UseXmmI2D = false; 513 } 514 } 515 if( FLAG_IS_DEFAULT(UseSSE42Intrinsics) ) { 516 if( supports_sse4_2() && UseSSE >= 4 ) { 517 UseSSE42Intrinsics = true; 518 } 519 } 520 521 // Use count leading zeros count instruction if available. 522 if (supports_lzcnt()) { 523 if (FLAG_IS_DEFAULT(UseCountLeadingZerosInstruction)) { 524 UseCountLeadingZerosInstruction = true; 525 } 526 } 527 528 // some defaults for AMD family 15h 529 if ( cpu_family() == 0x15 ) { 530 // On family 15h processors default is no sw prefetch 531 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) { 532 AllocatePrefetchStyle = 0; 533 } 534 // Also, if some other prefetch style is specified, default instruction type is PREFETCHW 535 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr)) { 536 AllocatePrefetchInstr = 3; 537 } 538 // On family 15h processors use XMM and UnalignedLoadStores for Array Copy 539 if( FLAG_IS_DEFAULT(UseXMMForArrayCopy) ) { 540 UseXMMForArrayCopy = true; 541 } 542 if( FLAG_IS_DEFAULT(UseUnalignedLoadStores) && UseXMMForArrayCopy ) { 543 UseUnalignedLoadStores = true; 544 } 545 } 546 547 } 548 549 if( is_intel() ) { // Intel cpus specific settings 550 if( FLAG_IS_DEFAULT(UseStoreImmI16) ) { 551 UseStoreImmI16 = false; // don't use it on Intel cpus 552 } 553 if( cpu_family() == 6 || cpu_family() == 15 ) { 554 if( FLAG_IS_DEFAULT(UseAddressNop) ) { 555 // Use it on all Intel cpus starting from PentiumPro 556 UseAddressNop = true; 557 } 558 } 559 if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) { 560 UseXmmLoadAndClearUpper = true; // use movsd on all Intel cpus 561 } 562 if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) { 563 if( supports_sse3() ) { 564 UseXmmRegToRegMoveAll = true; // use movaps, movapd on new Intel cpus 565 } else { 566 UseXmmRegToRegMoveAll = false; 567 } 568 } 569 if( cpu_family() == 6 && supports_sse3() ) { // New Intel cpus 570 #ifdef COMPILER2 571 if( FLAG_IS_DEFAULT(MaxLoopPad) ) { 572 // For new Intel cpus do the next optimization: 573 // don't align the beginning of a loop if there are enough instructions 574 // left (NumberOfLoopInstrToAlign defined in c2_globals.hpp) 575 // in current fetch line (OptoLoopAlignment) or the padding 576 // is big (> MaxLoopPad). 577 // Set MaxLoopPad to 11 for new Intel cpus to reduce number of 578 // generated NOP instructions. 11 is the largest size of one 579 // address NOP instruction '0F 1F' (see Assembler::nop(i)). 580 MaxLoopPad = 11; 581 } 582 #endif // COMPILER2 583 if( FLAG_IS_DEFAULT(UseXMMForArrayCopy) ) { 584 UseXMMForArrayCopy = true; // use SSE2 movq on new Intel cpus 585 } 586 if( supports_sse4_2() && supports_ht() ) { // Newest Intel cpus 587 if( FLAG_IS_DEFAULT(UseUnalignedLoadStores) && UseXMMForArrayCopy ) { 588 UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus 589 } 590 } 591 if( supports_sse4_2() && UseSSE >= 4 ) { 592 if( FLAG_IS_DEFAULT(UseSSE42Intrinsics)) { 593 UseSSE42Intrinsics = true; 594 } 595 } 596 } 597 } 598 599 // Use population count instruction if available. 600 if (supports_popcnt()) { 601 if (FLAG_IS_DEFAULT(UsePopCountInstruction)) { 602 UsePopCountInstruction = true; 603 } 604 } else if (UsePopCountInstruction) { 605 warning("POPCNT instruction is not available on this CPU"); 606 FLAG_SET_DEFAULT(UsePopCountInstruction, false); 607 } 608 609 #ifdef COMPILER2 610 if (UseFPUForSpilling) { 611 if (UseSSE < 2) { 612 // Only supported with SSE2+ 613 FLAG_SET_DEFAULT(UseFPUForSpilling, false); 614 } 615 } 616 #endif 617 618 assert(0 <= ReadPrefetchInstr && ReadPrefetchInstr <= 3, "invalid value"); 619 assert(0 <= AllocatePrefetchInstr && AllocatePrefetchInstr <= 3, "invalid value"); 620 621 // set valid Prefetch instruction 622 if( ReadPrefetchInstr < 0 ) ReadPrefetchInstr = 0; 623 if( ReadPrefetchInstr > 3 ) ReadPrefetchInstr = 3; 624 if( ReadPrefetchInstr == 3 && !supports_3dnow_prefetch() ) ReadPrefetchInstr = 0; 625 if( !supports_sse() && supports_3dnow_prefetch() ) ReadPrefetchInstr = 3; 626 627 if( AllocatePrefetchInstr < 0 ) AllocatePrefetchInstr = 0; 628 if( AllocatePrefetchInstr > 3 ) AllocatePrefetchInstr = 3; 629 if( AllocatePrefetchInstr == 3 && !supports_3dnow_prefetch() ) AllocatePrefetchInstr=0; 630 if( !supports_sse() && supports_3dnow_prefetch() ) AllocatePrefetchInstr = 3; 631 632 // Allocation prefetch settings 633 intx cache_line_size = prefetch_data_size(); 634 if( cache_line_size > AllocatePrefetchStepSize ) 635 AllocatePrefetchStepSize = cache_line_size; 636 637 assert(AllocatePrefetchLines > 0, "invalid value"); 638 if( AllocatePrefetchLines < 1 ) // set valid value in product VM 639 AllocatePrefetchLines = 3; 640 assert(AllocateInstancePrefetchLines > 0, "invalid value"); 641 if( AllocateInstancePrefetchLines < 1 ) // set valid value in product VM 642 AllocateInstancePrefetchLines = 1; 643 644 AllocatePrefetchDistance = allocate_prefetch_distance(); 645 AllocatePrefetchStyle = allocate_prefetch_style(); 646 647 if( is_intel() && cpu_family() == 6 && supports_sse3() ) { 648 if( AllocatePrefetchStyle == 2 ) { // watermark prefetching on Core 649 #ifdef _LP64 650 AllocatePrefetchDistance = 384; 651 #else 652 AllocatePrefetchDistance = 320; 653 #endif 654 } 655 if( supports_sse4_2() && supports_ht() ) { // Nehalem based cpus 656 AllocatePrefetchDistance = 192; 657 AllocatePrefetchLines = 4; 658 #ifdef COMPILER2 659 if (AggressiveOpts && FLAG_IS_DEFAULT(UseFPUForSpilling)) { 660 FLAG_SET_DEFAULT(UseFPUForSpilling, true); 661 } 662 #endif 663 } 664 } 665 assert(AllocatePrefetchDistance % AllocatePrefetchStepSize == 0, "invalid value"); 666 667 #ifdef _LP64 668 // Prefetch settings 669 PrefetchCopyIntervalInBytes = prefetch_copy_interval_in_bytes(); 670 PrefetchScanIntervalInBytes = prefetch_scan_interval_in_bytes(); 671 PrefetchFieldsAhead = prefetch_fields_ahead(); 672 #endif 673 674 #ifndef PRODUCT 675 if (PrintMiscellaneous && Verbose) { 676 tty->print_cr("Logical CPUs per core: %u", 677 logical_processors_per_package()); 678 tty->print("UseSSE=%d",UseSSE); 679 if (UseAVX > 0) { 680 tty->print(" UseAVX=%d",UseAVX); 681 } 682 tty->cr(); 683 tty->print("Allocation"); 684 if (AllocatePrefetchStyle <= 0 || UseSSE == 0 && !supports_3dnow_prefetch()) { 685 tty->print_cr(": no prefetching"); 686 } else { 687 tty->print(" prefetching: "); 688 if (UseSSE == 0 && supports_3dnow_prefetch()) { 689 tty->print("PREFETCHW"); 690 } else if (UseSSE >= 1) { 691 if (AllocatePrefetchInstr == 0) { 692 tty->print("PREFETCHNTA"); 693 } else if (AllocatePrefetchInstr == 1) { 694 tty->print("PREFETCHT0"); 695 } else if (AllocatePrefetchInstr == 2) { 696 tty->print("PREFETCHT2"); 697 } else if (AllocatePrefetchInstr == 3) { 698 tty->print("PREFETCHW"); 699 } 700 } 701 if (AllocatePrefetchLines > 1) { 702 tty->print_cr(" at distance %d, %d lines of %d bytes", AllocatePrefetchDistance, AllocatePrefetchLines, AllocatePrefetchStepSize); 703 } else { 704 tty->print_cr(" at distance %d, one line of %d bytes", AllocatePrefetchDistance, AllocatePrefetchStepSize); 705 } 706 } 707 708 if (PrefetchCopyIntervalInBytes > 0) { 709 tty->print_cr("PrefetchCopyIntervalInBytes %d", PrefetchCopyIntervalInBytes); 710 } 711 if (PrefetchScanIntervalInBytes > 0) { 712 tty->print_cr("PrefetchScanIntervalInBytes %d", PrefetchScanIntervalInBytes); 713 } 714 if (PrefetchFieldsAhead > 0) { 715 tty->print_cr("PrefetchFieldsAhead %d", PrefetchFieldsAhead); 716 } 717 } 718 #endif // !PRODUCT 719 } 720 721 void VM_Version::initialize() { 722 ResourceMark rm; 723 // Making this stub must be FIRST use of assembler 724 725 stub_blob = BufferBlob::create("getPsrInfo_stub", stub_size); 726 if (stub_blob == NULL) { 727 vm_exit_during_initialization("Unable to allocate getPsrInfo_stub"); 728 } 729 CodeBuffer c(stub_blob); 730 VM_Version_StubGenerator g(&c); 731 getPsrInfo_stub = CAST_TO_FN_PTR(getPsrInfo_stub_t, 732 g.generate_getPsrInfo()); 733 734 get_processor_features(); 735 }