1 /* 2 * Copyright (c) 1997, 2015, 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 "asm/macroAssembler.hpp" 27 #include "asm/macroAssembler.inline.hpp" 28 #include "memory/resourceArea.hpp" 29 #include "runtime/java.hpp" 30 #include "runtime/os.hpp" 31 #include "runtime/stubCodeGenerator.hpp" 32 #include "vm_version_x86.hpp" 33 34 35 int VM_Version::_cpu; 36 int VM_Version::_model; 37 int VM_Version::_stepping; 38 VM_Version::CpuidInfo VM_Version::_cpuid_info = { 0, }; 39 40 // Address of instruction which causes SEGV 41 address VM_Version::_cpuinfo_segv_addr = 0; 42 // Address of instruction after the one which causes SEGV 43 address VM_Version::_cpuinfo_cont_addr = 0; 44 45 static BufferBlob* stub_blob; 46 static const int stub_size = 1000; 47 48 extern "C" { 49 typedef void (*get_cpu_info_stub_t)(void*); 50 } 51 static get_cpu_info_stub_t get_cpu_info_stub = NULL; 52 53 54 class VM_Version_StubGenerator: public StubCodeGenerator { 55 public: 56 57 VM_Version_StubGenerator(CodeBuffer *c) : StubCodeGenerator(c) {} 58 59 address generate_get_cpu_info() { 60 // Flags to test CPU type. 61 const uint32_t HS_EFL_AC = 0x40000; 62 const uint32_t HS_EFL_ID = 0x200000; 63 // Values for when we don't have a CPUID instruction. 64 const int CPU_FAMILY_SHIFT = 8; 65 const uint32_t CPU_FAMILY_386 = (3 << CPU_FAMILY_SHIFT); 66 const uint32_t CPU_FAMILY_486 = (4 << CPU_FAMILY_SHIFT); 67 68 Label detect_486, cpu486, detect_586, std_cpuid1, std_cpuid4; 69 Label sef_cpuid, ext_cpuid, ext_cpuid1, ext_cpuid5, ext_cpuid7, done, wrapup; 70 Label legacy_setup, save_restore_except, legacy_save_restore, start_simd_check; 71 72 StubCodeMark mark(this, "VM_Version", "get_cpu_info_stub"); 73 # define __ _masm-> 74 75 address start = __ pc(); 76 77 // 78 // void get_cpu_info(VM_Version::CpuidInfo* cpuid_info); 79 // 80 // LP64: rcx and rdx are first and second argument registers on windows 81 82 __ push(rbp); 83 #ifdef _LP64 84 __ mov(rbp, c_rarg0); // cpuid_info address 85 #else 86 __ movptr(rbp, Address(rsp, 8)); // cpuid_info address 87 #endif 88 __ push(rbx); 89 __ push(rsi); 90 __ pushf(); // preserve rbx, and flags 91 __ pop(rax); 92 __ push(rax); 93 __ mov(rcx, rax); 94 // 95 // if we are unable to change the AC flag, we have a 386 96 // 97 __ xorl(rax, HS_EFL_AC); 98 __ push(rax); 99 __ popf(); 100 __ pushf(); 101 __ pop(rax); 102 __ cmpptr(rax, rcx); 103 __ jccb(Assembler::notEqual, detect_486); 104 105 __ movl(rax, CPU_FAMILY_386); 106 __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax); 107 __ jmp(done); 108 109 // 110 // If we are unable to change the ID flag, we have a 486 which does 111 // not support the "cpuid" instruction. 112 // 113 __ bind(detect_486); 114 __ mov(rax, rcx); 115 __ xorl(rax, HS_EFL_ID); 116 __ push(rax); 117 __ popf(); 118 __ pushf(); 119 __ pop(rax); 120 __ cmpptr(rcx, rax); 121 __ jccb(Assembler::notEqual, detect_586); 122 123 __ bind(cpu486); 124 __ movl(rax, CPU_FAMILY_486); 125 __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax); 126 __ jmp(done); 127 128 // 129 // At this point, we have a chip which supports the "cpuid" instruction 130 // 131 __ bind(detect_586); 132 __ xorl(rax, rax); 133 __ cpuid(); 134 __ orl(rax, rax); 135 __ jcc(Assembler::equal, cpu486); // if cpuid doesn't support an input 136 // value of at least 1, we give up and 137 // assume a 486 138 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); 139 __ movl(Address(rsi, 0), rax); 140 __ movl(Address(rsi, 4), rbx); 141 __ movl(Address(rsi, 8), rcx); 142 __ movl(Address(rsi,12), rdx); 143 144 __ cmpl(rax, 0xa); // Is cpuid(0xB) supported? 145 __ jccb(Assembler::belowEqual, std_cpuid4); 146 147 // 148 // cpuid(0xB) Processor Topology 149 // 150 __ movl(rax, 0xb); 151 __ xorl(rcx, rcx); // Threads level 152 __ cpuid(); 153 154 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB0_offset()))); 155 __ movl(Address(rsi, 0), rax); 156 __ movl(Address(rsi, 4), rbx); 157 __ movl(Address(rsi, 8), rcx); 158 __ movl(Address(rsi,12), rdx); 159 160 __ movl(rax, 0xb); 161 __ movl(rcx, 1); // Cores level 162 __ cpuid(); 163 __ push(rax); 164 __ andl(rax, 0x1f); // Determine if valid topology level 165 __ orl(rax, rbx); // eax[4:0] | ebx[0:15] == 0 indicates invalid level 166 __ andl(rax, 0xffff); 167 __ pop(rax); 168 __ jccb(Assembler::equal, std_cpuid4); 169 170 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB1_offset()))); 171 __ movl(Address(rsi, 0), rax); 172 __ movl(Address(rsi, 4), rbx); 173 __ movl(Address(rsi, 8), rcx); 174 __ movl(Address(rsi,12), rdx); 175 176 __ movl(rax, 0xb); 177 __ movl(rcx, 2); // Packages level 178 __ cpuid(); 179 __ push(rax); 180 __ andl(rax, 0x1f); // Determine if valid topology level 181 __ orl(rax, rbx); // eax[4:0] | ebx[0:15] == 0 indicates invalid level 182 __ andl(rax, 0xffff); 183 __ pop(rax); 184 __ jccb(Assembler::equal, std_cpuid4); 185 186 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB2_offset()))); 187 __ movl(Address(rsi, 0), rax); 188 __ movl(Address(rsi, 4), rbx); 189 __ movl(Address(rsi, 8), rcx); 190 __ movl(Address(rsi,12), rdx); 191 192 // 193 // cpuid(0x4) Deterministic cache params 194 // 195 __ bind(std_cpuid4); 196 __ movl(rax, 4); 197 __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x4) supported? 198 __ jccb(Assembler::greater, std_cpuid1); 199 200 __ xorl(rcx, rcx); // L1 cache 201 __ cpuid(); 202 __ push(rax); 203 __ andl(rax, 0x1f); // Determine if valid cache parameters used 204 __ orl(rax, rax); // eax[4:0] == 0 indicates invalid cache 205 __ pop(rax); 206 __ jccb(Assembler::equal, std_cpuid1); 207 208 __ lea(rsi, Address(rbp, in_bytes(VM_Version::dcp_cpuid4_offset()))); 209 __ movl(Address(rsi, 0), rax); 210 __ movl(Address(rsi, 4), rbx); 211 __ movl(Address(rsi, 8), rcx); 212 __ movl(Address(rsi,12), rdx); 213 214 // 215 // Standard cpuid(0x1) 216 // 217 __ bind(std_cpuid1); 218 __ movl(rax, 1); 219 __ cpuid(); 220 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset()))); 221 __ movl(Address(rsi, 0), rax); 222 __ movl(Address(rsi, 4), rbx); 223 __ movl(Address(rsi, 8), rcx); 224 __ movl(Address(rsi,12), rdx); 225 226 // 227 // Check if OS has enabled XGETBV instruction to access XCR0 228 // (OSXSAVE feature flag) and CPU supports AVX 229 // 230 __ andl(rcx, 0x18000000); // cpuid1 bits osxsave | avx 231 __ cmpl(rcx, 0x18000000); 232 __ jccb(Assembler::notEqual, sef_cpuid); // jump if AVX is not supported 233 234 // 235 // XCR0, XFEATURE_ENABLED_MASK register 236 // 237 __ xorl(rcx, rcx); // zero for XCR0 register 238 __ xgetbv(); 239 __ lea(rsi, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); 240 __ movl(Address(rsi, 0), rax); 241 __ movl(Address(rsi, 4), rdx); 242 243 // 244 // cpuid(0x7) Structured Extended Features 245 // 246 __ bind(sef_cpuid); 247 __ movl(rax, 7); 248 __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x7) supported? 249 __ jccb(Assembler::greater, ext_cpuid); 250 251 __ xorl(rcx, rcx); 252 __ cpuid(); 253 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset()))); 254 __ movl(Address(rsi, 0), rax); 255 __ movl(Address(rsi, 4), rbx); 256 257 // 258 // Extended cpuid(0x80000000) 259 // 260 __ bind(ext_cpuid); 261 __ movl(rax, 0x80000000); 262 __ cpuid(); 263 __ cmpl(rax, 0x80000000); // Is cpuid(0x80000001) supported? 264 __ jcc(Assembler::belowEqual, done); 265 __ cmpl(rax, 0x80000004); // Is cpuid(0x80000005) supported? 266 __ jccb(Assembler::belowEqual, ext_cpuid1); 267 __ cmpl(rax, 0x80000006); // Is cpuid(0x80000007) supported? 268 __ jccb(Assembler::belowEqual, ext_cpuid5); 269 __ cmpl(rax, 0x80000007); // Is cpuid(0x80000008) supported? 270 __ jccb(Assembler::belowEqual, ext_cpuid7); 271 // 272 // Extended cpuid(0x80000008) 273 // 274 __ movl(rax, 0x80000008); 275 __ cpuid(); 276 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid8_offset()))); 277 __ movl(Address(rsi, 0), rax); 278 __ movl(Address(rsi, 4), rbx); 279 __ movl(Address(rsi, 8), rcx); 280 __ movl(Address(rsi,12), rdx); 281 282 // 283 // Extended cpuid(0x80000007) 284 // 285 __ bind(ext_cpuid7); 286 __ movl(rax, 0x80000007); 287 __ cpuid(); 288 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid7_offset()))); 289 __ movl(Address(rsi, 0), rax); 290 __ movl(Address(rsi, 4), rbx); 291 __ movl(Address(rsi, 8), rcx); 292 __ movl(Address(rsi,12), rdx); 293 294 // 295 // Extended cpuid(0x80000005) 296 // 297 __ bind(ext_cpuid5); 298 __ movl(rax, 0x80000005); 299 __ cpuid(); 300 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid5_offset()))); 301 __ movl(Address(rsi, 0), rax); 302 __ movl(Address(rsi, 4), rbx); 303 __ movl(Address(rsi, 8), rcx); 304 __ movl(Address(rsi,12), rdx); 305 306 // 307 // Extended cpuid(0x80000001) 308 // 309 __ bind(ext_cpuid1); 310 __ movl(rax, 0x80000001); 311 __ cpuid(); 312 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid1_offset()))); 313 __ movl(Address(rsi, 0), rax); 314 __ movl(Address(rsi, 4), rbx); 315 __ movl(Address(rsi, 8), rcx); 316 __ movl(Address(rsi,12), rdx); 317 318 // 319 // Check if OS has enabled XGETBV instruction to access XCR0 320 // (OSXSAVE feature flag) and CPU supports AVX 321 // 322 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset()))); 323 __ movl(rcx, 0x18000000); // cpuid1 bits osxsave | avx 324 __ andl(rcx, Address(rsi, 8)); // cpuid1 bits osxsave | avx 325 __ cmpl(rcx, 0x18000000); 326 __ jccb(Assembler::notEqual, done); // jump if AVX is not supported 327 328 __ movl(rax, 0x6); 329 __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits sse | ymm 330 __ cmpl(rax, 0x6); 331 __ jccb(Assembler::equal, start_simd_check); // return if AVX is not supported 332 333 // we need to bridge farther than imm8, so we use this island as a thunk 334 __ bind(done); 335 __ jmp(wrapup); 336 337 __ bind(start_simd_check); 338 // 339 // Some OSs have a bug when upper 128/256bits of YMM/ZMM 340 // registers are not restored after a signal processing. 341 // Generate SEGV here (reference through NULL) 342 // and check upper YMM/ZMM bits after it. 343 // 344 intx saved_useavx = UseAVX; 345 intx saved_usesse = UseSSE; 346 // check _cpuid_info.sef_cpuid7_ebx.bits.avx512f 347 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset()))); 348 __ movl(rax, 0x10000); 349 __ andl(rax, Address(rsi, 4)); // xcr0 bits sse | ymm 350 __ cmpl(rax, 0x10000); 351 __ jccb(Assembler::notEqual, legacy_setup); // jump if EVEX is not supported 352 // check _cpuid_info.xem_xcr0_eax.bits.opmask 353 // check _cpuid_info.xem_xcr0_eax.bits.zmm512 354 // check _cpuid_info.xem_xcr0_eax.bits.zmm32 355 __ movl(rax, 0xE0); 356 __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits sse | ymm 357 __ cmpl(rax, 0xE0); 358 __ jccb(Assembler::notEqual, legacy_setup); // jump if EVEX is not supported 359 360 // EVEX setup: run in lowest evex mode 361 VM_Version::set_evex_cpuFeatures(); // Enable temporary to pass asserts 362 UseAVX = 3; 363 UseSSE = 2; 364 // load value into all 64 bytes of zmm7 register 365 __ movl(rcx, VM_Version::ymm_test_value()); 366 __ movdl(xmm0, rcx); 367 __ movl(rcx, 0xffff); 368 __ kmovwl(k1, rcx); 369 __ evpbroadcastd(xmm0, xmm0, Assembler::AVX_512bit); 370 __ evmovdqul(xmm7, xmm0, Assembler::AVX_512bit); 371 #ifdef _LP64 372 __ evmovdqul(xmm8, xmm0, Assembler::AVX_512bit); 373 __ evmovdqul(xmm31, xmm0, Assembler::AVX_512bit); 374 #endif 375 VM_Version::clean_cpuFeatures(); 376 __ jmp(save_restore_except); 377 378 __ bind(legacy_setup); 379 // AVX setup 380 VM_Version::set_avx_cpuFeatures(); // Enable temporary to pass asserts 381 UseAVX = 1; 382 UseSSE = 2; 383 // load value into all 32 bytes of ymm7 register 384 __ movl(rcx, VM_Version::ymm_test_value()); 385 386 __ movdl(xmm0, rcx); 387 __ pshufd(xmm0, xmm0, 0x00); 388 __ vinsertf128_high(xmm0, xmm0); 389 __ vmovdqu(xmm7, xmm0); 390 #ifdef _LP64 391 __ vmovdqu(xmm8, xmm0); 392 __ vmovdqu(xmm15, xmm0); 393 #endif 394 VM_Version::clean_cpuFeatures(); 395 396 __ bind(save_restore_except); 397 __ xorl(rsi, rsi); 398 VM_Version::set_cpuinfo_segv_addr(__ pc()); 399 // Generate SEGV 400 __ movl(rax, Address(rsi, 0)); 401 402 VM_Version::set_cpuinfo_cont_addr(__ pc()); 403 // Returns here after signal. Save xmm0 to check it later. 404 405 // check _cpuid_info.sef_cpuid7_ebx.bits.avx512f 406 __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset()))); 407 __ movl(rax, 0x10000); 408 __ andl(rax, Address(rsi, 4)); 409 __ cmpl(rax, 0x10000); 410 __ jccb(Assembler::notEqual, legacy_save_restore); 411 // check _cpuid_info.xem_xcr0_eax.bits.opmask 412 // check _cpuid_info.xem_xcr0_eax.bits.zmm512 413 // check _cpuid_info.xem_xcr0_eax.bits.zmm32 414 __ movl(rax, 0xE0); 415 __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits sse | ymm 416 __ cmpl(rax, 0xE0); 417 __ jccb(Assembler::notEqual, legacy_save_restore); 418 419 // EVEX check: run in lowest evex mode 420 VM_Version::set_evex_cpuFeatures(); // Enable temporary to pass asserts 421 UseAVX = 3; 422 UseSSE = 2; 423 __ lea(rsi, Address(rbp, in_bytes(VM_Version::zmm_save_offset()))); 424 __ evmovdqul(Address(rsi, 0), xmm0, Assembler::AVX_512bit); 425 __ evmovdqul(Address(rsi, 64), xmm7, Assembler::AVX_512bit); 426 #ifdef _LP64 427 __ evmovdqul(Address(rsi, 128), xmm8, Assembler::AVX_512bit); 428 __ evmovdqul(Address(rsi, 192), xmm31, Assembler::AVX_512bit); 429 #endif 430 VM_Version::clean_cpuFeatures(); 431 UseAVX = saved_useavx; 432 UseSSE = saved_usesse; 433 __ jmp(wrapup); 434 435 __ bind(legacy_save_restore); 436 // AVX check 437 VM_Version::set_avx_cpuFeatures(); // Enable temporary to pass asserts 438 UseAVX = 1; 439 UseSSE = 2; 440 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ymm_save_offset()))); 441 __ vmovdqu(Address(rsi, 0), xmm0); 442 __ vmovdqu(Address(rsi, 32), xmm7); 443 #ifdef _LP64 444 __ vmovdqu(Address(rsi, 64), xmm8); 445 __ vmovdqu(Address(rsi, 96), xmm15); 446 #endif 447 VM_Version::clean_cpuFeatures(); 448 UseAVX = saved_useavx; 449 UseSSE = saved_usesse; 450 451 __ bind(wrapup); 452 __ popf(); 453 __ pop(rsi); 454 __ pop(rbx); 455 __ pop(rbp); 456 __ ret(0); 457 458 # undef __ 459 460 return start; 461 }; 462 }; 463 464 void VM_Version::get_processor_features() { 465 466 _cpu = 4; // 486 by default 467 _model = 0; 468 _stepping = 0; 469 _features = 0; 470 _logical_processors_per_package = 1; 471 // i486 internal cache is both I&D and has a 16-byte line size 472 _L1_data_cache_line_size = 16; 473 474 // Get raw processor info 475 476 get_cpu_info_stub(&_cpuid_info); 477 478 assert_is_initialized(); 479 _cpu = extended_cpu_family(); 480 _model = extended_cpu_model(); 481 _stepping = cpu_stepping(); 482 483 if (cpu_family() > 4) { // it supports CPUID 484 _features = feature_flags(); 485 // Logical processors are only available on P4s and above, 486 // and only if hyperthreading is available. 487 _logical_processors_per_package = logical_processor_count(); 488 _L1_data_cache_line_size = L1_line_size(); 489 } 490 491 _supports_cx8 = supports_cmpxchg8(); 492 // xchg and xadd instructions 493 _supports_atomic_getset4 = true; 494 _supports_atomic_getadd4 = true; 495 LP64_ONLY(_supports_atomic_getset8 = true); 496 LP64_ONLY(_supports_atomic_getadd8 = true); 497 498 #ifdef _LP64 499 // OS should support SSE for x64 and hardware should support at least SSE2. 500 if (!VM_Version::supports_sse2()) { 501 vm_exit_during_initialization("Unknown x64 processor: SSE2 not supported"); 502 } 503 // in 64 bit the use of SSE2 is the minimum 504 if (UseSSE < 2) UseSSE = 2; 505 #endif 506 507 #ifdef AMD64 508 // flush_icache_stub have to be generated first. 509 // That is why Icache line size is hard coded in ICache class, 510 // see icache_x86.hpp. It is also the reason why we can't use 511 // clflush instruction in 32-bit VM since it could be running 512 // on CPU which does not support it. 513 // 514 // The only thing we can do is to verify that flushed 515 // ICache::line_size has correct value. 516 guarantee(_cpuid_info.std_cpuid1_edx.bits.clflush != 0, "clflush is not supported"); 517 // clflush_size is size in quadwords (8 bytes). 518 guarantee(_cpuid_info.std_cpuid1_ebx.bits.clflush_size == 8, "such clflush size is not supported"); 519 #endif 520 521 // If the OS doesn't support SSE, we can't use this feature even if the HW does 522 if (!os::supports_sse()) 523 _features &= ~(CPU_SSE|CPU_SSE2|CPU_SSE3|CPU_SSSE3|CPU_SSE4A|CPU_SSE4_1|CPU_SSE4_2); 524 525 if (UseSSE < 4) { 526 _features &= ~CPU_SSE4_1; 527 _features &= ~CPU_SSE4_2; 528 } 529 530 if (UseSSE < 3) { 531 _features &= ~CPU_SSE3; 532 _features &= ~CPU_SSSE3; 533 _features &= ~CPU_SSE4A; 534 } 535 536 if (UseSSE < 2) 537 _features &= ~CPU_SSE2; 538 539 if (UseSSE < 1) 540 _features &= ~CPU_SSE; 541 542 // first try initial setting and detect what we can support 543 if (UseAVX > 0) { 544 if (UseAVX > 2 && supports_evex()) { 545 UseAVX = 3; 546 } else if (UseAVX > 1 && supports_avx2()) { 547 UseAVX = 2; 548 } else if (UseAVX > 0 && supports_avx()) { 549 UseAVX = 1; 550 } else { 551 UseAVX = 0; 552 } 553 } else if (UseAVX < 0) { 554 UseAVX = 0; 555 } 556 557 if (UseAVX < 3) { 558 _features &= ~CPU_AVX512F; 559 _features &= ~CPU_AVX512DQ; 560 _features &= ~CPU_AVX512CD; 561 _features &= ~CPU_AVX512BW; 562 _features &= ~CPU_AVX512VL; 563 } 564 565 if (UseAVX < 2) 566 _features &= ~CPU_AVX2; 567 568 if (UseAVX < 1) 569 _features &= ~CPU_AVX; 570 571 if (!UseAES && !FLAG_IS_DEFAULT(UseAES)) 572 _features &= ~CPU_AES; 573 574 if (logical_processors_per_package() == 1) { 575 // HT processor could be installed on a system which doesn't support HT. 576 _features &= ~CPU_HT; 577 } 578 579 char buf[256]; 580 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%s%s%s%s%s%s%s%s", 581 cores_per_cpu(), threads_per_core(), 582 cpu_family(), _model, _stepping, 583 (supports_cmov() ? ", cmov" : ""), 584 (supports_cmpxchg8() ? ", cx8" : ""), 585 (supports_fxsr() ? ", fxsr" : ""), 586 (supports_mmx() ? ", mmx" : ""), 587 (supports_sse() ? ", sse" : ""), 588 (supports_sse2() ? ", sse2" : ""), 589 (supports_sse3() ? ", sse3" : ""), 590 (supports_ssse3()? ", ssse3": ""), 591 (supports_sse4_1() ? ", sse4.1" : ""), 592 (supports_sse4_2() ? ", sse4.2" : ""), 593 (supports_popcnt() ? ", popcnt" : ""), 594 (supports_avx() ? ", avx" : ""), 595 (supports_avx2() ? ", avx2" : ""), 596 (supports_aes() ? ", aes" : ""), 597 (supports_clmul() ? ", clmul" : ""), 598 (supports_erms() ? ", erms" : ""), 599 (supports_rtm() ? ", rtm" : ""), 600 (supports_mmx_ext() ? ", mmxext" : ""), 601 (supports_3dnow_prefetch() ? ", 3dnowpref" : ""), 602 (supports_lzcnt() ? ", lzcnt": ""), 603 (supports_sse4a() ? ", sse4a": ""), 604 (supports_ht() ? ", ht": ""), 605 (supports_tsc() ? ", tsc": ""), 606 (supports_tscinv_bit() ? ", tscinvbit": ""), 607 (supports_tscinv() ? ", tscinv": ""), 608 (supports_bmi1() ? ", bmi1" : ""), 609 (supports_bmi2() ? ", bmi2" : ""), 610 (supports_adx() ? ", adx" : ""), 611 (supports_evex() ? ", evex" : "")); 612 _features_string = os::strdup(buf); 613 614 // UseSSE is set to the smaller of what hardware supports and what 615 // the command line requires. I.e., you cannot set UseSSE to 2 on 616 // older Pentiums which do not support it. 617 if (UseSSE > 4) UseSSE=4; 618 if (UseSSE < 0) UseSSE=0; 619 if (!supports_sse4_1()) // Drop to 3 if no SSE4 support 620 UseSSE = MIN2((intx)3,UseSSE); 621 if (!supports_sse3()) // Drop to 2 if no SSE3 support 622 UseSSE = MIN2((intx)2,UseSSE); 623 if (!supports_sse2()) // Drop to 1 if no SSE2 support 624 UseSSE = MIN2((intx)1,UseSSE); 625 if (!supports_sse ()) // Drop to 0 if no SSE support 626 UseSSE = 0; 627 628 // Use AES instructions if available. 629 if (supports_aes()) { 630 if (FLAG_IS_DEFAULT(UseAES)) { 631 FLAG_SET_DEFAULT(UseAES, true); 632 } 633 if (!UseAES) { 634 if (UseAESIntrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) { 635 warning("AES intrinsics require UseAES flag to be enabled. Intrinsics will be disabled."); 636 } 637 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 638 } else { 639 if (UseSSE > 2) { 640 if (FLAG_IS_DEFAULT(UseAESIntrinsics)) { 641 FLAG_SET_DEFAULT(UseAESIntrinsics, true); 642 } 643 } else { 644 // The AES intrinsic stubs require AES instruction support (of course) 645 // but also require sse3 mode or higher for instructions it use. 646 if (UseAESIntrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) { 647 warning("X86 AES intrinsics require SSE3 instructions or higher. Intrinsics will be disabled."); 648 } 649 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 650 } 651 652 // --AES-CTR begins-- 653 if (!UseAESIntrinsics) { 654 if (UseAESCTRIntrinsics && !FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) { 655 warning("AES-CTR intrinsics require UseAESIntrinsics flag to be enabled. Intrinsics will be disabled."); 656 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false); 657 } 658 } else { 659 if(supports_sse4_1() && UseSSE >= 4) { 660 if (FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) { 661 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, true); 662 } 663 } else { 664 // The AES-CTR intrinsic stubs require AES instruction support (of course) 665 // but also require sse4.1 mode or higher for instructions it use. 666 if (UseAESCTRIntrinsics && !FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) { 667 warning("X86 AES-CTR intrinsics require SSE4.1 instructions or higher. Intrinsics will be disabled."); 668 } 669 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false); 670 } 671 } 672 // --AES-CTR ends-- 673 } 674 } else if (UseAES || UseAESIntrinsics || UseAESCTRIntrinsics) { 675 if (UseAES && !FLAG_IS_DEFAULT(UseAES)) { 676 warning("AES instructions are not available on this CPU"); 677 FLAG_SET_DEFAULT(UseAES, false); 678 } 679 if (UseAESIntrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) { 680 warning("AES intrinsics are not available on this CPU"); 681 FLAG_SET_DEFAULT(UseAESIntrinsics, false); 682 } 683 if (UseAESCTRIntrinsics && !FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) { 684 warning("AES-CTR intrinsics are not available on this CPU"); 685 FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false); 686 } 687 } 688 689 // Use CLMUL instructions if available. 690 if (supports_clmul()) { 691 if (FLAG_IS_DEFAULT(UseCLMUL)) { 692 UseCLMUL = true; 693 } 694 } else if (UseCLMUL) { 695 if (!FLAG_IS_DEFAULT(UseCLMUL)) 696 warning("CLMUL instructions not available on this CPU (AVX may also be required)"); 697 FLAG_SET_DEFAULT(UseCLMUL, false); 698 } 699 700 if (UseCLMUL && (UseSSE > 2)) { 701 if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) { 702 UseCRC32Intrinsics = true; 703 } 704 } else if (UseCRC32Intrinsics) { 705 if (!FLAG_IS_DEFAULT(UseCRC32Intrinsics)) 706 warning("CRC32 Intrinsics requires CLMUL instructions (not available on this CPU)"); 707 FLAG_SET_DEFAULT(UseCRC32Intrinsics, false); 708 } 709 710 if (supports_sse4_2()) { 711 if (FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) { 712 UseCRC32CIntrinsics = true; 713 } 714 } 715 else if (UseCRC32CIntrinsics) { 716 if (!FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) { 717 warning("CRC32C intrinsics are not available on this CPU"); 718 } 719 FLAG_SET_DEFAULT(UseCRC32CIntrinsics, false); 720 } 721 722 // GHASH/GCM intrinsics 723 if (UseCLMUL && (UseSSE > 2)) { 724 if (FLAG_IS_DEFAULT(UseGHASHIntrinsics)) { 725 UseGHASHIntrinsics = true; 726 } 727 } else if (UseGHASHIntrinsics) { 728 if (!FLAG_IS_DEFAULT(UseGHASHIntrinsics)) 729 warning("GHASH intrinsic requires CLMUL and SSE2 instructions on this CPU"); 730 FLAG_SET_DEFAULT(UseGHASHIntrinsics, false); 731 } 732 733 if (UseSHA) { 734 warning("SHA instructions are not available on this CPU"); 735 FLAG_SET_DEFAULT(UseSHA, false); 736 } 737 738 if (UseSHA1Intrinsics) { 739 warning("Intrinsics for SHA-1 crypto hash functions not available on this CPU."); 740 FLAG_SET_DEFAULT(UseSHA1Intrinsics, false); 741 } 742 743 if (UseSHA256Intrinsics) { 744 warning("Intrinsics for SHA-224 and SHA-256 crypto hash functions not available on this CPU."); 745 FLAG_SET_DEFAULT(UseSHA256Intrinsics, false); 746 } 747 748 if (UseSHA512Intrinsics) { 749 warning("Intrinsics for SHA-384 and SHA-512 crypto hash functions not available on this CPU."); 750 FLAG_SET_DEFAULT(UseSHA512Intrinsics, false); 751 } 752 753 if (UseAdler32Intrinsics) { 754 warning("Adler32Intrinsics not available on this CPU."); 755 FLAG_SET_DEFAULT(UseAdler32Intrinsics, false); 756 } 757 758 // Adjust RTM (Restricted Transactional Memory) flags 759 if (!supports_rtm() && UseRTMLocking) { 760 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 761 // setting during arguments processing. See use_biased_locking(). 762 // VM_Version_init() is executed after UseBiasedLocking is used 763 // in Thread::allocate(). 764 vm_exit_during_initialization("RTM instructions are not available on this CPU"); 765 } 766 767 #if INCLUDE_RTM_OPT 768 if (UseRTMLocking) { 769 if (is_intel_family_core()) { 770 if ((_model == CPU_MODEL_HASWELL_E3) || 771 (_model == CPU_MODEL_HASWELL_E7 && _stepping < 3) || 772 (_model == CPU_MODEL_BROADWELL && _stepping < 4)) { 773 // currently a collision between SKL and HSW_E3 774 if (!UnlockExperimentalVMOptions && UseAVX < 3) { 775 vm_exit_during_initialization("UseRTMLocking is only available as experimental option on this platform. It must be enabled via -XX:+UnlockExperimentalVMOptions flag."); 776 } else { 777 warning("UseRTMLocking is only available as experimental option on this platform."); 778 } 779 } 780 } 781 if (!FLAG_IS_CMDLINE(UseRTMLocking)) { 782 // RTM locking should be used only for applications with 783 // high lock contention. For now we do not use it by default. 784 vm_exit_during_initialization("UseRTMLocking flag should be only set on command line"); 785 } 786 if (!is_power_of_2(RTMTotalCountIncrRate)) { 787 warning("RTMTotalCountIncrRate must be a power of 2, resetting it to 64"); 788 FLAG_SET_DEFAULT(RTMTotalCountIncrRate, 64); 789 } 790 if (RTMAbortRatio < 0 || RTMAbortRatio > 100) { 791 warning("RTMAbortRatio must be in the range 0 to 100, resetting it to 50"); 792 FLAG_SET_DEFAULT(RTMAbortRatio, 50); 793 } 794 } else { // !UseRTMLocking 795 if (UseRTMForStackLocks) { 796 if (!FLAG_IS_DEFAULT(UseRTMForStackLocks)) { 797 warning("UseRTMForStackLocks flag should be off when UseRTMLocking flag is off"); 798 } 799 FLAG_SET_DEFAULT(UseRTMForStackLocks, false); 800 } 801 if (UseRTMDeopt) { 802 FLAG_SET_DEFAULT(UseRTMDeopt, false); 803 } 804 if (PrintPreciseRTMLockingStatistics) { 805 FLAG_SET_DEFAULT(PrintPreciseRTMLockingStatistics, false); 806 } 807 } 808 #else 809 if (UseRTMLocking) { 810 // Only C2 does RTM locking optimization. 811 // Can't continue because UseRTMLocking affects UseBiasedLocking flag 812 // setting during arguments processing. See use_biased_locking(). 813 vm_exit_during_initialization("RTM locking optimization is not supported in this VM"); 814 } 815 #endif 816 817 #ifdef COMPILER2 818 if (UseFPUForSpilling) { 819 if (UseSSE < 2) { 820 // Only supported with SSE2+ 821 FLAG_SET_DEFAULT(UseFPUForSpilling, false); 822 } 823 } 824 #endif 825 #if defined(COMPILER2) || INCLUDE_JVMCI 826 if (MaxVectorSize > 0) { 827 if (!is_power_of_2(MaxVectorSize)) { 828 warning("MaxVectorSize must be a power of 2"); 829 FLAG_SET_DEFAULT(MaxVectorSize, 64); 830 } 831 if (MaxVectorSize > 64) { 832 FLAG_SET_DEFAULT(MaxVectorSize, 64); 833 } 834 if (MaxVectorSize > 16 && (UseAVX == 0 || !os_supports_avx_vectors())) { 835 // 32 bytes vectors (in YMM) are only supported with AVX+ 836 FLAG_SET_DEFAULT(MaxVectorSize, 16); 837 } 838 if (UseSSE < 2) { 839 // Vectors (in XMM) are only supported with SSE2+ 840 FLAG_SET_DEFAULT(MaxVectorSize, 0); 841 } 842 #if defined(COMPILER2) && defined(ASSERT) 843 if (supports_avx() && PrintMiscellaneous && Verbose && TraceNewVectors) { 844 tty->print_cr("State of YMM registers after signal handle:"); 845 int nreg = 2 LP64_ONLY(+2); 846 const char* ymm_name[4] = {"0", "7", "8", "15"}; 847 for (int i = 0; i < nreg; i++) { 848 tty->print("YMM%s:", ymm_name[i]); 849 for (int j = 7; j >=0; j--) { 850 tty->print(" %x", _cpuid_info.ymm_save[i*8 + j]); 851 } 852 tty->cr(); 853 } 854 } 855 #endif // COMPILER2 && ASSERT 856 } 857 #endif // COMPILER2 || INCLUDE_JVMCI 858 859 #ifdef COMPILER2 860 #ifdef _LP64 861 if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) { 862 UseMultiplyToLenIntrinsic = true; 863 } 864 if (FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) { 865 UseSquareToLenIntrinsic = true; 866 } 867 if (FLAG_IS_DEFAULT(UseMulAddIntrinsic)) { 868 UseMulAddIntrinsic = true; 869 } 870 if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) { 871 UseMontgomeryMultiplyIntrinsic = true; 872 } 873 if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) { 874 UseMontgomerySquareIntrinsic = true; 875 } 876 #else 877 if (UseMultiplyToLenIntrinsic) { 878 if (!FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) { 879 warning("multiplyToLen intrinsic is not available in 32-bit VM"); 880 } 881 FLAG_SET_DEFAULT(UseMultiplyToLenIntrinsic, false); 882 } 883 if (UseMontgomeryMultiplyIntrinsic) { 884 if (!FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) { 885 warning("montgomeryMultiply intrinsic is not available in 32-bit VM"); 886 } 887 FLAG_SET_DEFAULT(UseMontgomeryMultiplyIntrinsic, false); 888 } 889 if (UseMontgomerySquareIntrinsic) { 890 if (!FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) { 891 warning("montgomerySquare intrinsic is not available in 32-bit VM"); 892 } 893 FLAG_SET_DEFAULT(UseMontgomerySquareIntrinsic, false); 894 } 895 if (UseSquareToLenIntrinsic) { 896 if (!FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) { 897 warning("squareToLen intrinsic is not available in 32-bit VM"); 898 } 899 FLAG_SET_DEFAULT(UseSquareToLenIntrinsic, false); 900 } 901 if (UseMulAddIntrinsic) { 902 if (!FLAG_IS_DEFAULT(UseMulAddIntrinsic)) { 903 warning("mulAdd intrinsic is not available in 32-bit VM"); 904 } 905 FLAG_SET_DEFAULT(UseMulAddIntrinsic, false); 906 } 907 #endif 908 #endif // COMPILER2 909 910 // On new cpus instructions which update whole XMM register should be used 911 // to prevent partial register stall due to dependencies on high half. 912 // 913 // UseXmmLoadAndClearUpper == true --> movsd(xmm, mem) 914 // UseXmmLoadAndClearUpper == false --> movlpd(xmm, mem) 915 // UseXmmRegToRegMoveAll == true --> movaps(xmm, xmm), movapd(xmm, xmm). 916 // UseXmmRegToRegMoveAll == false --> movss(xmm, xmm), movsd(xmm, xmm). 917 918 if( is_amd() ) { // AMD cpus specific settings 919 if( supports_sse2() && FLAG_IS_DEFAULT(UseAddressNop) ) { 920 // Use it on new AMD cpus starting from Opteron. 921 UseAddressNop = true; 922 } 923 if( supports_sse2() && FLAG_IS_DEFAULT(UseNewLongLShift) ) { 924 // Use it on new AMD cpus starting from Opteron. 925 UseNewLongLShift = true; 926 } 927 if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) { 928 if (supports_sse4a()) { 929 UseXmmLoadAndClearUpper = true; // use movsd only on '10h' Opteron 930 } else { 931 UseXmmLoadAndClearUpper = false; 932 } 933 } 934 if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) { 935 if( supports_sse4a() ) { 936 UseXmmRegToRegMoveAll = true; // use movaps, movapd only on '10h' 937 } else { 938 UseXmmRegToRegMoveAll = false; 939 } 940 } 941 if( FLAG_IS_DEFAULT(UseXmmI2F) ) { 942 if( supports_sse4a() ) { 943 UseXmmI2F = true; 944 } else { 945 UseXmmI2F = false; 946 } 947 } 948 if( FLAG_IS_DEFAULT(UseXmmI2D) ) { 949 if( supports_sse4a() ) { 950 UseXmmI2D = true; 951 } else { 952 UseXmmI2D = false; 953 } 954 } 955 if (supports_sse4_2() && UseSSE >= 4) { 956 if (FLAG_IS_DEFAULT(UseSSE42Intrinsics)) { 957 FLAG_SET_DEFAULT(UseSSE42Intrinsics, true); 958 } 959 } else { 960 if (UseSSE42Intrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) { 961 warning("SSE4.2 intrinsics require SSE4.2 instructions or higher. Intrinsics will be disabled."); 962 } 963 FLAG_SET_DEFAULT(UseSSE42Intrinsics, false); 964 } 965 966 // some defaults for AMD family 15h 967 if ( cpu_family() == 0x15 ) { 968 // On family 15h processors default is no sw prefetch 969 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) { 970 AllocatePrefetchStyle = 0; 971 } 972 // Also, if some other prefetch style is specified, default instruction type is PREFETCHW 973 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr)) { 974 AllocatePrefetchInstr = 3; 975 } 976 // On family 15h processors use XMM and UnalignedLoadStores for Array Copy 977 if (supports_sse2() && FLAG_IS_DEFAULT(UseXMMForArrayCopy)) { 978 UseXMMForArrayCopy = true; 979 } 980 if (supports_sse2() && FLAG_IS_DEFAULT(UseUnalignedLoadStores)) { 981 UseUnalignedLoadStores = true; 982 } 983 } 984 985 #ifdef COMPILER2 986 if (MaxVectorSize > 16) { 987 // Limit vectors size to 16 bytes on current AMD cpus. 988 FLAG_SET_DEFAULT(MaxVectorSize, 16); 989 } 990 #endif // COMPILER2 991 } 992 993 if( is_intel() ) { // Intel cpus specific settings 994 if( FLAG_IS_DEFAULT(UseStoreImmI16) ) { 995 UseStoreImmI16 = false; // don't use it on Intel cpus 996 } 997 if( cpu_family() == 6 || cpu_family() == 15 ) { 998 if( FLAG_IS_DEFAULT(UseAddressNop) ) { 999 // Use it on all Intel cpus starting from PentiumPro 1000 UseAddressNop = true; 1001 } 1002 } 1003 if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) { 1004 UseXmmLoadAndClearUpper = true; // use movsd on all Intel cpus 1005 } 1006 if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) { 1007 if( supports_sse3() ) { 1008 UseXmmRegToRegMoveAll = true; // use movaps, movapd on new Intel cpus 1009 } else { 1010 UseXmmRegToRegMoveAll = false; 1011 } 1012 } 1013 if( cpu_family() == 6 && supports_sse3() ) { // New Intel cpus 1014 #ifdef COMPILER2 1015 if( FLAG_IS_DEFAULT(MaxLoopPad) ) { 1016 // For new Intel cpus do the next optimization: 1017 // don't align the beginning of a loop if there are enough instructions 1018 // left (NumberOfLoopInstrToAlign defined in c2_globals.hpp) 1019 // in current fetch line (OptoLoopAlignment) or the padding 1020 // is big (> MaxLoopPad). 1021 // Set MaxLoopPad to 11 for new Intel cpus to reduce number of 1022 // generated NOP instructions. 11 is the largest size of one 1023 // address NOP instruction '0F 1F' (see Assembler::nop(i)). 1024 MaxLoopPad = 11; 1025 } 1026 #endif // COMPILER2 1027 if (FLAG_IS_DEFAULT(UseXMMForArrayCopy)) { 1028 UseXMMForArrayCopy = true; // use SSE2 movq on new Intel cpus 1029 } 1030 if (supports_sse4_2() && supports_ht()) { // Newest Intel cpus 1031 if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) { 1032 UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus 1033 } 1034 } 1035 if (supports_sse4_2() && UseSSE >= 4) { 1036 if (FLAG_IS_DEFAULT(UseSSE42Intrinsics)) { 1037 FLAG_SET_DEFAULT(UseSSE42Intrinsics, true); 1038 } 1039 } else { 1040 if (UseSSE42Intrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) { 1041 warning("SSE4.2 intrinsics require SSE4.2 instructions or higher. Intrinsics will be disabled."); 1042 } 1043 FLAG_SET_DEFAULT(UseSSE42Intrinsics, false); 1044 } 1045 } 1046 if ((cpu_family() == 0x06) && 1047 ((extended_cpu_model() == 0x36) || // Centerton 1048 (extended_cpu_model() == 0x37) || // Silvermont 1049 (extended_cpu_model() == 0x4D))) { 1050 #ifdef COMPILER2 1051 if (FLAG_IS_DEFAULT(OptoScheduling)) { 1052 OptoScheduling = true; 1053 } 1054 #endif 1055 if (supports_sse4_2()) { // Silvermont 1056 if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) { 1057 UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus 1058 } 1059 } 1060 } 1061 if(FLAG_IS_DEFAULT(AllocatePrefetchInstr) && supports_3dnow_prefetch()) { 1062 AllocatePrefetchInstr = 3; 1063 } 1064 } 1065 1066 #ifdef _LP64 1067 if (UseSSE42Intrinsics) { 1068 if (FLAG_IS_DEFAULT(UseVectorizedMismatchIntrinsic)) { 1069 UseVectorizedMismatchIntrinsic = true; 1070 } 1071 } else if (UseVectorizedMismatchIntrinsic) { 1072 if (!FLAG_IS_DEFAULT(UseVectorizedMismatchIntrinsic)) 1073 warning("vectorizedMismatch intrinsics are not available on this CPU"); 1074 FLAG_SET_DEFAULT(UseVectorizedMismatchIntrinsic, false); 1075 } 1076 #else 1077 if (UseVectorizedMismatchIntrinsic) { 1078 if (!FLAG_IS_DEFAULT(UseVectorizedMismatchIntrinsic)) { 1079 warning("vectorizedMismatch intrinsic is not available in 32-bit VM"); 1080 } 1081 FLAG_SET_DEFAULT(UseVectorizedMismatchIntrinsic, false); 1082 } 1083 #endif // _LP64 1084 1085 // Use count leading zeros count instruction if available. 1086 if (supports_lzcnt()) { 1087 if (FLAG_IS_DEFAULT(UseCountLeadingZerosInstruction)) { 1088 UseCountLeadingZerosInstruction = true; 1089 } 1090 } else if (UseCountLeadingZerosInstruction) { 1091 warning("lzcnt instruction is not available on this CPU"); 1092 FLAG_SET_DEFAULT(UseCountLeadingZerosInstruction, false); 1093 } 1094 1095 // Use count trailing zeros instruction if available 1096 if (supports_bmi1()) { 1097 // tzcnt does not require VEX prefix 1098 if (FLAG_IS_DEFAULT(UseCountTrailingZerosInstruction)) { 1099 if (!UseBMI1Instructions && !FLAG_IS_DEFAULT(UseBMI1Instructions)) { 1100 // Don't use tzcnt if BMI1 is switched off on command line. 1101 UseCountTrailingZerosInstruction = false; 1102 } else { 1103 UseCountTrailingZerosInstruction = true; 1104 } 1105 } 1106 } else if (UseCountTrailingZerosInstruction) { 1107 warning("tzcnt instruction is not available on this CPU"); 1108 FLAG_SET_DEFAULT(UseCountTrailingZerosInstruction, false); 1109 } 1110 1111 // BMI instructions (except tzcnt) use an encoding with VEX prefix. 1112 // VEX prefix is generated only when AVX > 0. 1113 if (supports_bmi1() && supports_avx()) { 1114 if (FLAG_IS_DEFAULT(UseBMI1Instructions)) { 1115 UseBMI1Instructions = true; 1116 } 1117 } else if (UseBMI1Instructions) { 1118 warning("BMI1 instructions are not available on this CPU (AVX is also required)"); 1119 FLAG_SET_DEFAULT(UseBMI1Instructions, false); 1120 } 1121 1122 if (supports_bmi2() && supports_avx()) { 1123 if (FLAG_IS_DEFAULT(UseBMI2Instructions)) { 1124 UseBMI2Instructions = true; 1125 } 1126 } else if (UseBMI2Instructions) { 1127 warning("BMI2 instructions are not available on this CPU (AVX is also required)"); 1128 FLAG_SET_DEFAULT(UseBMI2Instructions, false); 1129 } 1130 1131 // Use population count instruction if available. 1132 if (supports_popcnt()) { 1133 if (FLAG_IS_DEFAULT(UsePopCountInstruction)) { 1134 UsePopCountInstruction = true; 1135 } 1136 } else if (UsePopCountInstruction) { 1137 warning("POPCNT instruction is not available on this CPU"); 1138 FLAG_SET_DEFAULT(UsePopCountInstruction, false); 1139 } 1140 1141 // Use fast-string operations if available. 1142 if (supports_erms()) { 1143 if (FLAG_IS_DEFAULT(UseFastStosb)) { 1144 UseFastStosb = true; 1145 } 1146 } else if (UseFastStosb) { 1147 warning("fast-string operations are not available on this CPU"); 1148 FLAG_SET_DEFAULT(UseFastStosb, false); 1149 } 1150 1151 #ifdef COMPILER2 1152 if (FLAG_IS_DEFAULT(AlignVector)) { 1153 // Modern processors allow misaligned memory operations for vectors. 1154 AlignVector = !UseUnalignedLoadStores; 1155 } 1156 #endif // COMPILER2 1157 1158 if( AllocatePrefetchInstr == 3 && !supports_3dnow_prefetch() ) AllocatePrefetchInstr=0; 1159 if( !supports_sse() && supports_3dnow_prefetch() ) AllocatePrefetchInstr = 3; 1160 1161 // Allocation prefetch settings 1162 intx cache_line_size = prefetch_data_size(); 1163 if( cache_line_size > AllocatePrefetchStepSize ) 1164 AllocatePrefetchStepSize = cache_line_size; 1165 1166 AllocatePrefetchDistance = allocate_prefetch_distance(); 1167 AllocatePrefetchStyle = allocate_prefetch_style(); 1168 1169 if (is_intel() && cpu_family() == 6 && supports_sse3()) { 1170 if (AllocatePrefetchStyle == 2) { // watermark prefetching on Core 1171 #ifdef _LP64 1172 AllocatePrefetchDistance = 384; 1173 #else 1174 AllocatePrefetchDistance = 320; 1175 #endif 1176 } 1177 if (supports_sse4_2() && supports_ht()) { // Nehalem based cpus 1178 AllocatePrefetchDistance = 192; 1179 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) { 1180 FLAG_SET_DEFAULT(AllocatePrefetchLines, 4); 1181 } 1182 } 1183 #ifdef COMPILER2 1184 if (supports_sse4_2()) { 1185 if (FLAG_IS_DEFAULT(UseFPUForSpilling)) { 1186 FLAG_SET_DEFAULT(UseFPUForSpilling, true); 1187 } 1188 } 1189 #endif 1190 } 1191 1192 #ifdef _LP64 1193 // Prefetch settings 1194 PrefetchCopyIntervalInBytes = prefetch_copy_interval_in_bytes(); 1195 PrefetchScanIntervalInBytes = prefetch_scan_interval_in_bytes(); 1196 PrefetchFieldsAhead = prefetch_fields_ahead(); 1197 #endif 1198 1199 if (FLAG_IS_DEFAULT(ContendedPaddingWidth) && 1200 (cache_line_size > ContendedPaddingWidth)) 1201 ContendedPaddingWidth = cache_line_size; 1202 1203 // This machine allows unaligned memory accesses 1204 if (FLAG_IS_DEFAULT(UseUnalignedAccesses)) { 1205 FLAG_SET_DEFAULT(UseUnalignedAccesses, true); 1206 } 1207 1208 #ifndef PRODUCT 1209 if (PrintMiscellaneous && Verbose) { 1210 tty->print_cr("Logical CPUs per core: %u", 1211 logical_processors_per_package()); 1212 tty->print_cr("L1 data cache line size: %u", L1_data_cache_line_size()); 1213 tty->print("UseSSE=%d", (int) UseSSE); 1214 if (UseAVX > 0) { 1215 tty->print(" UseAVX=%d", (int) UseAVX); 1216 } 1217 if (UseAES) { 1218 tty->print(" UseAES=1"); 1219 } 1220 #ifdef COMPILER2 1221 if (MaxVectorSize > 0) { 1222 tty->print(" MaxVectorSize=%d", (int) MaxVectorSize); 1223 } 1224 #endif 1225 tty->cr(); 1226 tty->print("Allocation"); 1227 if (AllocatePrefetchStyle <= 0 || UseSSE == 0 && !supports_3dnow_prefetch()) { 1228 tty->print_cr(": no prefetching"); 1229 } else { 1230 tty->print(" prefetching: "); 1231 if (UseSSE == 0 && supports_3dnow_prefetch()) { 1232 tty->print("PREFETCHW"); 1233 } else if (UseSSE >= 1) { 1234 if (AllocatePrefetchInstr == 0) { 1235 tty->print("PREFETCHNTA"); 1236 } else if (AllocatePrefetchInstr == 1) { 1237 tty->print("PREFETCHT0"); 1238 } else if (AllocatePrefetchInstr == 2) { 1239 tty->print("PREFETCHT2"); 1240 } else if (AllocatePrefetchInstr == 3) { 1241 tty->print("PREFETCHW"); 1242 } 1243 } 1244 if (AllocatePrefetchLines > 1) { 1245 tty->print_cr(" at distance %d, %d lines of %d bytes", (int) AllocatePrefetchDistance, (int) AllocatePrefetchLines, (int) AllocatePrefetchStepSize); 1246 } else { 1247 tty->print_cr(" at distance %d, one line of %d bytes", (int) AllocatePrefetchDistance, (int) AllocatePrefetchStepSize); 1248 } 1249 } 1250 1251 if (PrefetchCopyIntervalInBytes > 0) { 1252 tty->print_cr("PrefetchCopyIntervalInBytes %d", (int) PrefetchCopyIntervalInBytes); 1253 } 1254 if (PrefetchScanIntervalInBytes > 0) { 1255 tty->print_cr("PrefetchScanIntervalInBytes %d", (int) PrefetchScanIntervalInBytes); 1256 } 1257 if (PrefetchFieldsAhead > 0) { 1258 tty->print_cr("PrefetchFieldsAhead %d", (int) PrefetchFieldsAhead); 1259 } 1260 if (ContendedPaddingWidth > 0) { 1261 tty->print_cr("ContendedPaddingWidth %d", (int) ContendedPaddingWidth); 1262 } 1263 } 1264 #endif // !PRODUCT 1265 } 1266 1267 bool VM_Version::use_biased_locking() { 1268 #if INCLUDE_RTM_OPT 1269 // RTM locking is most useful when there is high lock contention and 1270 // low data contention. With high lock contention the lock is usually 1271 // inflated and biased locking is not suitable for that case. 1272 // RTM locking code requires that biased locking is off. 1273 // Note: we can't switch off UseBiasedLocking in get_processor_features() 1274 // because it is used by Thread::allocate() which is called before 1275 // VM_Version::initialize(). 1276 if (UseRTMLocking && UseBiasedLocking) { 1277 if (FLAG_IS_DEFAULT(UseBiasedLocking)) { 1278 FLAG_SET_DEFAULT(UseBiasedLocking, false); 1279 } else { 1280 warning("Biased locking is not supported with RTM locking; ignoring UseBiasedLocking flag." ); 1281 UseBiasedLocking = false; 1282 } 1283 } 1284 #endif 1285 return UseBiasedLocking; 1286 } 1287 1288 void VM_Version::initialize() { 1289 ResourceMark rm; 1290 // Making this stub must be FIRST use of assembler 1291 1292 stub_blob = BufferBlob::create("get_cpu_info_stub", stub_size); 1293 if (stub_blob == NULL) { 1294 vm_exit_during_initialization("Unable to allocate get_cpu_info_stub"); 1295 } 1296 CodeBuffer c(stub_blob); 1297 VM_Version_StubGenerator g(&c); 1298 get_cpu_info_stub = CAST_TO_FN_PTR(get_cpu_info_stub_t, 1299 g.generate_get_cpu_info()); 1300 1301 get_processor_features(); 1302 }