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 uint64_t VM_Version::_cpuFeatures;
  39 const char*           VM_Version::_features_str = "";
  40 VM_Version::CpuidInfo VM_Version::_cpuid_info   = { 0, };
  41 
  42 // Address of instruction which causes SEGV
  43 address VM_Version::_cpuinfo_segv_addr = 0;
  44 // Address of instruction after the one which causes SEGV
  45 address VM_Version::_cpuinfo_cont_addr = 0;
  46 
  47 static BufferBlob* stub_blob;
  48 static const int stub_size = 1000;
  49 
  50 extern "C" {
  51   typedef void (*get_cpu_info_stub_t)(void*);
  52 }
  53 static get_cpu_info_stub_t get_cpu_info_stub = NULL;
  54 
  55 
  56 class VM_Version_StubGenerator: public StubCodeGenerator {
  57  public:
  58 
  59   VM_Version_StubGenerator(CodeBuffer *c) : StubCodeGenerator(c) {}
  60 
  61   address generate_get_cpu_info() {
  62     // Flags to test CPU type.
  63     const uint32_t HS_EFL_AC = 0x40000;
  64     const uint32_t HS_EFL_ID = 0x200000;
  65     // Values for when we don't have a CPUID instruction.
  66     const int      CPU_FAMILY_SHIFT = 8;
  67     const uint32_t CPU_FAMILY_386 = (3 << CPU_FAMILY_SHIFT);
  68     const uint32_t CPU_FAMILY_486 = (4 << CPU_FAMILY_SHIFT);
  69 
  70     Label detect_486, cpu486, detect_586, std_cpuid1, std_cpuid4;
  71     Label sef_cpuid, ext_cpuid, ext_cpuid1, ext_cpuid5, ext_cpuid7, done, wrapup;
  72     Label legacy_setup, save_restore_except, legacy_save_restore, start_simd_check;
  73 
  74     StubCodeMark mark(this, "VM_Version", "get_cpu_info_stub");
  75 #   define __ _masm->
  76 
  77     address start = __ pc();
  78 
  79     //
  80     // void get_cpu_info(VM_Version::CpuidInfo* cpuid_info);
  81     //
  82     // LP64: rcx and rdx are first and second argument registers on windows
  83 
  84     __ push(rbp);
  85 #ifdef _LP64
  86     __ mov(rbp, c_rarg0); // cpuid_info address
  87 #else
  88     __ movptr(rbp, Address(rsp, 8)); // cpuid_info address
  89 #endif
  90     __ push(rbx);
  91     __ push(rsi);
  92     __ pushf();          // preserve rbx, and flags
  93     __ pop(rax);
  94     __ push(rax);
  95     __ mov(rcx, rax);
  96     //
  97     // if we are unable to change the AC flag, we have a 386
  98     //
  99     __ xorl(rax, HS_EFL_AC);
 100     __ push(rax);
 101     __ popf();
 102     __ pushf();
 103     __ pop(rax);
 104     __ cmpptr(rax, rcx);
 105     __ jccb(Assembler::notEqual, detect_486);
 106 
 107     __ movl(rax, CPU_FAMILY_386);
 108     __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax);
 109     __ jmp(done);
 110 
 111     //
 112     // If we are unable to change the ID flag, we have a 486 which does
 113     // not support the "cpuid" instruction.
 114     //
 115     __ bind(detect_486);
 116     __ mov(rax, rcx);
 117     __ xorl(rax, HS_EFL_ID);
 118     __ push(rax);
 119     __ popf();
 120     __ pushf();
 121     __ pop(rax);
 122     __ cmpptr(rcx, rax);
 123     __ jccb(Assembler::notEqual, detect_586);
 124 
 125     __ bind(cpu486);
 126     __ movl(rax, CPU_FAMILY_486);
 127     __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax);
 128     __ jmp(done);
 129 
 130     //
 131     // At this point, we have a chip which supports the "cpuid" instruction
 132     //
 133     __ bind(detect_586);
 134     __ xorl(rax, rax);
 135     __ cpuid();
 136     __ orl(rax, rax);
 137     __ jcc(Assembler::equal, cpu486);   // if cpuid doesn't support an input
 138                                         // value of at least 1, we give up and
 139                                         // assume a 486
 140     __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset())));
 141     __ movl(Address(rsi, 0), rax);
 142     __ movl(Address(rsi, 4), rbx);
 143     __ movl(Address(rsi, 8), rcx);
 144     __ movl(Address(rsi,12), rdx);
 145 
 146     __ cmpl(rax, 0xa);                  // Is cpuid(0xB) supported?
 147     __ jccb(Assembler::belowEqual, std_cpuid4);
 148 
 149     //
 150     // cpuid(0xB) Processor Topology
 151     //
 152     __ movl(rax, 0xb);
 153     __ xorl(rcx, rcx);   // Threads level
 154     __ cpuid();
 155 
 156     __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB0_offset())));
 157     __ movl(Address(rsi, 0), rax);
 158     __ movl(Address(rsi, 4), rbx);
 159     __ movl(Address(rsi, 8), rcx);
 160     __ movl(Address(rsi,12), rdx);
 161 
 162     __ movl(rax, 0xb);
 163     __ movl(rcx, 1);     // Cores level
 164     __ cpuid();
 165     __ push(rax);
 166     __ andl(rax, 0x1f);  // Determine if valid topology level
 167     __ orl(rax, rbx);    // eax[4:0] | ebx[0:15] == 0 indicates invalid level
 168     __ andl(rax, 0xffff);
 169     __ pop(rax);
 170     __ jccb(Assembler::equal, std_cpuid4);
 171 
 172     __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB1_offset())));
 173     __ movl(Address(rsi, 0), rax);
 174     __ movl(Address(rsi, 4), rbx);
 175     __ movl(Address(rsi, 8), rcx);
 176     __ movl(Address(rsi,12), rdx);
 177 
 178     __ movl(rax, 0xb);
 179     __ movl(rcx, 2);     // Packages level
 180     __ cpuid();
 181     __ push(rax);
 182     __ andl(rax, 0x1f);  // Determine if valid topology level
 183     __ orl(rax, rbx);    // eax[4:0] | ebx[0:15] == 0 indicates invalid level
 184     __ andl(rax, 0xffff);
 185     __ pop(rax);
 186     __ jccb(Assembler::equal, std_cpuid4);
 187 
 188     __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB2_offset())));
 189     __ movl(Address(rsi, 0), rax);
 190     __ movl(Address(rsi, 4), rbx);
 191     __ movl(Address(rsi, 8), rcx);
 192     __ movl(Address(rsi,12), rdx);
 193 
 194     //
 195     // cpuid(0x4) Deterministic cache params
 196     //
 197     __ bind(std_cpuid4);
 198     __ movl(rax, 4);
 199     __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x4) supported?
 200     __ jccb(Assembler::greater, std_cpuid1);
 201 
 202     __ xorl(rcx, rcx);   // L1 cache
 203     __ cpuid();
 204     __ push(rax);
 205     __ andl(rax, 0x1f);  // Determine if valid cache parameters used
 206     __ orl(rax, rax);    // eax[4:0] == 0 indicates invalid cache
 207     __ pop(rax);
 208     __ jccb(Assembler::equal, std_cpuid1);
 209 
 210     __ lea(rsi, Address(rbp, in_bytes(VM_Version::dcp_cpuid4_offset())));
 211     __ movl(Address(rsi, 0), rax);
 212     __ movl(Address(rsi, 4), rbx);
 213     __ movl(Address(rsi, 8), rcx);
 214     __ movl(Address(rsi,12), rdx);
 215 
 216     //
 217     // Standard cpuid(0x1)
 218     //
 219     __ bind(std_cpuid1);
 220     __ movl(rax, 1);
 221     __ cpuid();
 222     __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
 223     __ movl(Address(rsi, 0), rax);
 224     __ movl(Address(rsi, 4), rbx);
 225     __ movl(Address(rsi, 8), rcx);
 226     __ movl(Address(rsi,12), rdx);
 227 
 228     //
 229     // Check if OS has enabled XGETBV instruction to access XCR0
 230     // (OSXSAVE feature flag) and CPU supports AVX
 231     //
 232     __ andl(rcx, 0x18000000); // cpuid1 bits osxsave | avx
 233     __ cmpl(rcx, 0x18000000);
 234     __ jccb(Assembler::notEqual, sef_cpuid); // jump if AVX is not supported
 235 
 236     //
 237     // XCR0, XFEATURE_ENABLED_MASK register
 238     //
 239     __ xorl(rcx, rcx);   // zero for XCR0 register
 240     __ xgetbv();
 241     __ lea(rsi, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset())));
 242     __ movl(Address(rsi, 0), rax);
 243     __ movl(Address(rsi, 4), rdx);
 244 
 245     //
 246     // cpuid(0x7) Structured Extended Features
 247     //
 248     __ bind(sef_cpuid);
 249     __ movl(rax, 7);
 250     __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x7) supported?
 251     __ jccb(Assembler::greater, ext_cpuid);
 252 
 253     __ xorl(rcx, rcx);
 254     __ cpuid();
 255     __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset())));
 256     __ movl(Address(rsi, 0), rax);
 257     __ movl(Address(rsi, 4), rbx);
 258 
 259     //
 260     // Extended cpuid(0x80000000)
 261     //
 262     __ bind(ext_cpuid);
 263     __ movl(rax, 0x80000000);
 264     __ cpuid();
 265     __ cmpl(rax, 0x80000000);     // Is cpuid(0x80000001) supported?
 266     __ jcc(Assembler::belowEqual, done);
 267     __ cmpl(rax, 0x80000004);     // Is cpuid(0x80000005) supported?
 268     __ jccb(Assembler::belowEqual, ext_cpuid1);
 269     __ cmpl(rax, 0x80000006);     // Is cpuid(0x80000007) supported?
 270     __ jccb(Assembler::belowEqual, ext_cpuid5);
 271     __ cmpl(rax, 0x80000007);     // Is cpuid(0x80000008) supported?
 272     __ jccb(Assembler::belowEqual, ext_cpuid7);
 273     //
 274     // Extended cpuid(0x80000008)
 275     //
 276     __ movl(rax, 0x80000008);
 277     __ cpuid();
 278     __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid8_offset())));
 279     __ movl(Address(rsi, 0), rax);
 280     __ movl(Address(rsi, 4), rbx);
 281     __ movl(Address(rsi, 8), rcx);
 282     __ movl(Address(rsi,12), rdx);
 283 
 284     //
 285     // Extended cpuid(0x80000007)
 286     //
 287     __ bind(ext_cpuid7);
 288     __ movl(rax, 0x80000007);
 289     __ cpuid();
 290     __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid7_offset())));
 291     __ movl(Address(rsi, 0), rax);
 292     __ movl(Address(rsi, 4), rbx);
 293     __ movl(Address(rsi, 8), rcx);
 294     __ movl(Address(rsi,12), rdx);
 295 
 296     //
 297     // Extended cpuid(0x80000005)
 298     //
 299     __ bind(ext_cpuid5);
 300     __ movl(rax, 0x80000005);
 301     __ cpuid();
 302     __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid5_offset())));
 303     __ movl(Address(rsi, 0), rax);
 304     __ movl(Address(rsi, 4), rbx);
 305     __ movl(Address(rsi, 8), rcx);
 306     __ movl(Address(rsi,12), rdx);
 307 
 308     //
 309     // Extended cpuid(0x80000001)
 310     //
 311     __ bind(ext_cpuid1);
 312     __ movl(rax, 0x80000001);
 313     __ cpuid();
 314     __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid1_offset())));
 315     __ movl(Address(rsi, 0), rax);
 316     __ movl(Address(rsi, 4), rbx);
 317     __ movl(Address(rsi, 8), rcx);
 318     __ movl(Address(rsi,12), rdx);
 319 
 320     //
 321     // Check if OS has enabled XGETBV instruction to access XCR0
 322     // (OSXSAVE feature flag) and CPU supports AVX
 323     //
 324     __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
 325     __ movl(rcx, 0x18000000); // cpuid1 bits osxsave | avx
 326     __ andl(rcx, Address(rsi, 8)); // cpuid1 bits osxsave | avx
 327     __ cmpl(rcx, 0x18000000);
 328     __ jccb(Assembler::notEqual, done); // jump if AVX is not supported
 329 
 330     __ movl(rax, 0x6);
 331     __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits sse | ymm
 332     __ cmpl(rax, 0x6);
 333     __ jccb(Assembler::equal, start_simd_check); // return if AVX is not supported
 334 
 335     // we need to bridge farther than imm8, so we use this island as a thunk
 336     __ bind(done);
 337     __ jmp(wrapup);
 338 
 339     __ bind(start_simd_check);
 340     //
 341     // Some OSs have a bug when upper 128/256bits of YMM/ZMM
 342     // registers are not restored after a signal processing.
 343     // Generate SEGV here (reference through NULL)
 344     // and check upper YMM/ZMM bits after it.
 345     //
 346     intx saved_useavx = UseAVX;
 347     intx saved_usesse = UseSSE;
 348     // check _cpuid_info.sef_cpuid7_ebx.bits.avx512f
 349     __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset())));
 350     __ movl(rax, 0x10000);
 351     __ andl(rax, Address(rsi, 4)); // xcr0 bits sse | ymm
 352     __ cmpl(rax, 0x10000);
 353     __ jccb(Assembler::notEqual, legacy_setup); // jump if EVEX is not supported
 354     // check _cpuid_info.xem_xcr0_eax.bits.opmask
 355     // check _cpuid_info.xem_xcr0_eax.bits.zmm512
 356     // check _cpuid_info.xem_xcr0_eax.bits.zmm32
 357     __ movl(rax, 0xE0);
 358     __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits sse | ymm
 359     __ cmpl(rax, 0xE0);
 360     __ jccb(Assembler::notEqual, legacy_setup); // jump if EVEX is not supported
 361 
 362     // EVEX setup: run in lowest evex mode
 363     VM_Version::set_evex_cpuFeatures(); // Enable temporary to pass asserts
 364     UseAVX = 3;
 365     UseSSE = 2;
 366     // load value into all 64 bytes of zmm7 register
 367     __ movl(rcx, VM_Version::ymm_test_value());
 368     __ movdl(xmm0, rcx);
 369     __ movl(rcx, 0xffff);
 370     __ kmovwl(k1, rcx);
 371     __ evpbroadcastd(xmm0, xmm0, Assembler::AVX_512bit);
 372     __ evmovdqul(xmm7, xmm0, Assembler::AVX_512bit);
 373 #ifdef _LP64
 374     __ evmovdqul(xmm8, xmm0, Assembler::AVX_512bit);
 375     __ evmovdqul(xmm31, xmm0, Assembler::AVX_512bit);
 376 #endif
 377     VM_Version::clean_cpuFeatures();
 378     __ jmp(save_restore_except);
 379 
 380     __ bind(legacy_setup);
 381     // AVX setup
 382     VM_Version::set_avx_cpuFeatures(); // Enable temporary to pass asserts
 383     UseAVX = 1;
 384     UseSSE = 2;
 385     // load value into all 32 bytes of ymm7 register
 386     __ movl(rcx, VM_Version::ymm_test_value());
 387 
 388     __ movdl(xmm0, rcx);
 389     __ pshufd(xmm0, xmm0, 0x00);
 390     __ vinsertf128h(xmm0, xmm0, xmm0);
 391     __ vmovdqu(xmm7, xmm0);
 392 #ifdef _LP64
 393     __ vmovdqu(xmm8, xmm0);
 394     __ vmovdqu(xmm15, xmm0);
 395 #endif
 396     VM_Version::clean_cpuFeatures();
 397 
 398     __ bind(save_restore_except);
 399     __ xorl(rsi, rsi);
 400     VM_Version::set_cpuinfo_segv_addr(__ pc());
 401     // Generate SEGV
 402     __ movl(rax, Address(rsi, 0));
 403 
 404     VM_Version::set_cpuinfo_cont_addr(__ pc());
 405     // Returns here after signal. Save xmm0 to check it later.
 406 
 407     // check _cpuid_info.sef_cpuid7_ebx.bits.avx512f
 408     __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset())));
 409     __ movl(rax, 0x10000);
 410     __ andl(rax, Address(rsi, 4));
 411     __ cmpl(rax, 0x10000);
 412     __ jccb(Assembler::notEqual, legacy_save_restore);
 413     // check _cpuid_info.xem_xcr0_eax.bits.opmask
 414     // check _cpuid_info.xem_xcr0_eax.bits.zmm512
 415     // check _cpuid_info.xem_xcr0_eax.bits.zmm32
 416     __ movl(rax, 0xE0);
 417     __ andl(rax, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset()))); // xcr0 bits sse | ymm
 418     __ cmpl(rax, 0xE0);
 419     __ jccb(Assembler::notEqual, legacy_save_restore);
 420 
 421     // EVEX check: run in lowest evex mode
 422     VM_Version::set_evex_cpuFeatures(); // Enable temporary to pass asserts
 423     UseAVX = 3;
 424     UseSSE = 2;
 425     __ lea(rsi, Address(rbp, in_bytes(VM_Version::zmm_save_offset())));
 426     __ evmovdqul(Address(rsi, 0), xmm0, Assembler::AVX_512bit);
 427     __ evmovdqul(Address(rsi, 64), xmm7, Assembler::AVX_512bit);
 428 #ifdef _LP64
 429     __ evmovdqul(Address(rsi, 128), xmm8, Assembler::AVX_512bit);
 430     __ evmovdqul(Address(rsi, 192), xmm31, Assembler::AVX_512bit);
 431 #endif
 432     VM_Version::clean_cpuFeatures();
 433     UseAVX = saved_useavx;
 434     UseSSE = saved_usesse;
 435     __ jmp(wrapup);
 436 
 437     __ bind(legacy_save_restore);
 438     // AVX check
 439     VM_Version::set_avx_cpuFeatures(); // Enable temporary to pass asserts
 440     UseAVX = 1;
 441     UseSSE = 2;
 442     __ lea(rsi, Address(rbp, in_bytes(VM_Version::ymm_save_offset())));
 443     __ vmovdqu(Address(rsi, 0), xmm0);
 444     __ vmovdqu(Address(rsi, 32), xmm7);
 445 #ifdef _LP64
 446     __ vmovdqu(Address(rsi, 64), xmm8);
 447     __ vmovdqu(Address(rsi, 96), xmm15);
 448 #endif
 449     VM_Version::clean_cpuFeatures();
 450     UseAVX = saved_useavx;
 451     UseSSE = saved_usesse;
 452 
 453     __ bind(wrapup);
 454     __ popf();
 455     __ pop(rsi);
 456     __ pop(rbx);
 457     __ pop(rbp);
 458     __ ret(0);
 459 
 460 #   undef __
 461 
 462     return start;
 463   };
 464 };
 465 
 466 void VM_Version::get_processor_features() {
 467 
 468   _cpu = 4; // 486 by default
 469   _model = 0;
 470   _stepping = 0;
 471   _cpuFeatures = 0;
 472   _logical_processors_per_package = 1;
 473   // i486 internal cache is both I&D and has a 16-byte line size
 474   _L1_data_cache_line_size = 16;
 475 
 476   if (!Use486InstrsOnly) {
 477     // Get raw processor info
 478 
 479     get_cpu_info_stub(&_cpuid_info);
 480 
 481     assert_is_initialized();
 482     _cpu = extended_cpu_family();
 483     _model = extended_cpu_model();
 484     _stepping = cpu_stepping();
 485 
 486     if (cpu_family() > 4) { // it supports CPUID
 487       _cpuFeatures = feature_flags();
 488       // Logical processors are only available on P4s and above,
 489       // and only if hyperthreading is available.
 490       _logical_processors_per_package = logical_processor_count();
 491       _L1_data_cache_line_size = L1_line_size();
 492     }
 493   }
 494 
 495   _supports_cx8 = supports_cmpxchg8();
 496   // xchg and xadd instructions
 497   _supports_atomic_getset4 = true;
 498   _supports_atomic_getadd4 = true;
 499   LP64_ONLY(_supports_atomic_getset8 = true);
 500   LP64_ONLY(_supports_atomic_getadd8 = true);
 501 
 502 #ifdef _LP64
 503   // OS should support SSE for x64 and hardware should support at least SSE2.
 504   if (!VM_Version::supports_sse2()) {
 505     vm_exit_during_initialization("Unknown x64 processor: SSE2 not supported");
 506   }
 507   // in 64 bit the use of SSE2 is the minimum
 508   if (UseSSE < 2) UseSSE = 2;
 509 #endif
 510 
 511 #ifdef AMD64
 512   // flush_icache_stub have to be generated first.
 513   // That is why Icache line size is hard coded in ICache class,
 514   // see icache_x86.hpp. It is also the reason why we can't use
 515   // clflush instruction in 32-bit VM since it could be running
 516   // on CPU which does not support it.
 517   //
 518   // The only thing we can do is to verify that flushed
 519   // ICache::line_size has correct value.
 520   guarantee(_cpuid_info.std_cpuid1_edx.bits.clflush != 0, "clflush is not supported");
 521   // clflush_size is size in quadwords (8 bytes).
 522   guarantee(_cpuid_info.std_cpuid1_ebx.bits.clflush_size == 8, "such clflush size is not supported");
 523 #endif
 524 
 525   // If the OS doesn't support SSE, we can't use this feature even if the HW does
 526   if (!os::supports_sse())
 527     _cpuFeatures &= ~(CPU_SSE|CPU_SSE2|CPU_SSE3|CPU_SSSE3|CPU_SSE4A|CPU_SSE4_1|CPU_SSE4_2);
 528 
 529   if (UseSSE < 4) {
 530     _cpuFeatures &= ~CPU_SSE4_1;
 531     _cpuFeatures &= ~CPU_SSE4_2;
 532   }
 533 
 534   if (UseSSE < 3) {
 535     _cpuFeatures &= ~CPU_SSE3;
 536     _cpuFeatures &= ~CPU_SSSE3;
 537     _cpuFeatures &= ~CPU_SSE4A;
 538   }
 539 
 540   if (UseSSE < 2)
 541     _cpuFeatures &= ~CPU_SSE2;
 542 
 543   if (UseSSE < 1)
 544     _cpuFeatures &= ~CPU_SSE;
 545 
 546   // first try initial setting and detect what we can support
 547   if (UseAVX > 0) {
 548     if (UseAVX > 2 && supports_evex()) {
 549       UseAVX = 3;
 550     } else if (UseAVX > 1 && supports_avx2()) {
 551       UseAVX = 2;
 552     } else if (UseAVX > 0 && supports_avx()) {
 553       UseAVX = 1;
 554     } else {
 555       UseAVX = 0;
 556     }
 557   } else if (UseAVX < 0) {
 558     UseAVX = 0;
 559   }
 560 
 561   if (UseAVX < 3) {
 562     _cpuFeatures &= ~CPU_AVX512F;
 563     _cpuFeatures &= ~CPU_AVX512DQ;
 564     _cpuFeatures &= ~CPU_AVX512CD;
 565     _cpuFeatures &= ~CPU_AVX512BW;
 566     _cpuFeatures &= ~CPU_AVX512VL;
 567   }
 568 
 569   if (UseAVX < 2)
 570     _cpuFeatures &= ~CPU_AVX2;
 571 
 572   if (UseAVX < 1)
 573     _cpuFeatures &= ~CPU_AVX;
 574 
 575   if (!UseAES && !FLAG_IS_DEFAULT(UseAES))
 576     _cpuFeatures &= ~CPU_AES;
 577 
 578   if (logical_processors_per_package() == 1) {
 579     // HT processor could be installed on a system which doesn't support HT.
 580     _cpuFeatures &= ~CPU_HT;
 581   }
 582 
 583   char buf[256];
 584   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",
 585                cores_per_cpu(), threads_per_core(),
 586                cpu_family(), _model, _stepping,
 587                (supports_cmov() ? ", cmov" : ""),
 588                (supports_cmpxchg8() ? ", cx8" : ""),
 589                (supports_fxsr() ? ", fxsr" : ""),
 590                (supports_mmx()  ? ", mmx"  : ""),
 591                (supports_sse()  ? ", sse"  : ""),
 592                (supports_sse2() ? ", sse2" : ""),
 593                (supports_sse3() ? ", sse3" : ""),
 594                (supports_ssse3()? ", ssse3": ""),
 595                (supports_sse4_1() ? ", sse4.1" : ""),
 596                (supports_sse4_2() ? ", sse4.2" : ""),
 597                (supports_popcnt() ? ", popcnt" : ""),
 598                (supports_avx()    ? ", avx" : ""),
 599                (supports_avx2()   ? ", avx2" : ""),
 600                (supports_aes()    ? ", aes" : ""),
 601                (supports_clmul()  ? ", clmul" : ""),
 602                (supports_erms()   ? ", erms" : ""),
 603                (supports_rtm()    ? ", rtm" : ""),
 604                (supports_mmx_ext() ? ", mmxext" : ""),
 605                (supports_3dnow_prefetch() ? ", 3dnowpref" : ""),
 606                (supports_lzcnt()   ? ", lzcnt": ""),
 607                (supports_sse4a()   ? ", sse4a": ""),
 608                (supports_ht() ? ", ht": ""),
 609                (supports_tsc() ? ", tsc": ""),
 610                (supports_tscinv_bit() ? ", tscinvbit": ""),
 611                (supports_tscinv() ? ", tscinv": ""),
 612                (supports_bmi1() ? ", bmi1" : ""),
 613                (supports_bmi2() ? ", bmi2" : ""),
 614                (supports_adx() ? ", adx" : ""),
 615                (supports_evex() ? ", evex" : ""));
 616   _features_str = os::strdup(buf);
 617 
 618   // UseSSE is set to the smaller of what hardware supports and what
 619   // the command line requires.  I.e., you cannot set UseSSE to 2 on
 620   // older Pentiums which do not support it.
 621   if (UseSSE > 4) UseSSE=4;
 622   if (UseSSE < 0) UseSSE=0;
 623   if (!supports_sse4_1()) // Drop to 3 if no SSE4 support
 624     UseSSE = MIN2((intx)3,UseSSE);
 625   if (!supports_sse3()) // Drop to 2 if no SSE3 support
 626     UseSSE = MIN2((intx)2,UseSSE);
 627   if (!supports_sse2()) // Drop to 1 if no SSE2 support
 628     UseSSE = MIN2((intx)1,UseSSE);
 629   if (!supports_sse ()) // Drop to 0 if no SSE  support
 630     UseSSE = 0;
 631 
 632   // Use AES instructions if available.
 633   if (supports_aes()) {
 634     if (FLAG_IS_DEFAULT(UseAES)) {
 635       UseAES = true;
 636     }
 637   } else if (UseAES) {
 638     if (!FLAG_IS_DEFAULT(UseAES))
 639       warning("AES instructions are not available on this CPU");
 640     FLAG_SET_DEFAULT(UseAES, false);
 641   }
 642 
 643   // Use CLMUL instructions if available.
 644   if (supports_clmul()) {
 645     if (FLAG_IS_DEFAULT(UseCLMUL)) {
 646       UseCLMUL = true;
 647     }
 648   } else if (UseCLMUL) {
 649     if (!FLAG_IS_DEFAULT(UseCLMUL))
 650       warning("CLMUL instructions not available on this CPU (AVX may also be required)");
 651     FLAG_SET_DEFAULT(UseCLMUL, false);
 652   }
 653 
 654   if (UseCLMUL && (UseSSE > 2)) {
 655     if (FLAG_IS_DEFAULT(UseCRC32Intrinsics)) {
 656       UseCRC32Intrinsics = true;
 657     }
 658   } else if (UseCRC32Intrinsics) {
 659     if (!FLAG_IS_DEFAULT(UseCRC32Intrinsics))
 660       warning("CRC32 Intrinsics requires CLMUL instructions (not available on this CPU)");
 661     FLAG_SET_DEFAULT(UseCRC32Intrinsics, false);
 662   }
 663 
 664   if (supports_sse4_2()) {
 665     if (FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) {
 666       UseCRC32CIntrinsics = true;
 667     }
 668   }
 669   else if (UseCRC32CIntrinsics) {
 670     if (!FLAG_IS_DEFAULT(UseCRC32CIntrinsics)) {
 671       warning("CRC32C intrinsics are not available on this CPU");
 672     }
 673     FLAG_SET_DEFAULT(UseCRC32CIntrinsics, false);
 674   }
 675 
 676   // The AES intrinsic stubs require AES instruction support (of course)
 677   // but also require sse3 mode for instructions it use.
 678   if (UseAES && (UseSSE > 2)) {
 679     if (FLAG_IS_DEFAULT(UseAESIntrinsics)) {
 680       UseAESIntrinsics = true;
 681     }
 682   } else if (UseAESIntrinsics) {
 683     if (!FLAG_IS_DEFAULT(UseAESIntrinsics))
 684       warning("AES intrinsics are not available on this CPU");
 685     FLAG_SET_DEFAULT(UseAESIntrinsics, false);
 686   }
 687 
 688   // GHASH/GCM intrinsics
 689   if (UseCLMUL && (UseSSE > 2)) {
 690     if (FLAG_IS_DEFAULT(UseGHASHIntrinsics)) {
 691       UseGHASHIntrinsics = true;
 692     }
 693   } else if (UseGHASHIntrinsics) {
 694     if (!FLAG_IS_DEFAULT(UseGHASHIntrinsics))
 695       warning("GHASH intrinsic requires CLMUL and SSE2 instructions on this CPU");
 696     FLAG_SET_DEFAULT(UseGHASHIntrinsics, false);
 697   }
 698 
 699   if (UseSHA) {
 700     warning("SHA instructions are not available on this CPU");
 701     FLAG_SET_DEFAULT(UseSHA, false);
 702   }
 703 
 704   if (UseSHA1Intrinsics) {
 705     warning("Intrinsics for SHA-1 crypto hash functions not available on this CPU.");
 706     FLAG_SET_DEFAULT(UseSHA1Intrinsics, false);
 707   }
 708 
 709   if (UseSHA256Intrinsics) {
 710     warning("Intrinsics for SHA-224 and SHA-256 crypto hash functions not available on this CPU.");
 711     FLAG_SET_DEFAULT(UseSHA256Intrinsics, false);
 712   }
 713 
 714   if (UseSHA512Intrinsics) {
 715     warning("Intrinsics for SHA-384 and SHA-512 crypto hash functions not available on this CPU.");
 716     FLAG_SET_DEFAULT(UseSHA512Intrinsics, false);
 717   }
 718 
 719   if (UseAdler32Intrinsics) {
 720     warning("Adler32Intrinsics not available on this CPU.");
 721     FLAG_SET_DEFAULT(UseAdler32Intrinsics, false);
 722   }
 723 
 724   // Adjust RTM (Restricted Transactional Memory) flags
 725   if (!supports_rtm() && UseRTMLocking) {
 726     // Can't continue because UseRTMLocking affects UseBiasedLocking flag
 727     // setting during arguments processing. See use_biased_locking().
 728     // VM_Version_init() is executed after UseBiasedLocking is used
 729     // in Thread::allocate().
 730     vm_exit_during_initialization("RTM instructions are not available on this CPU");
 731   }
 732 
 733 #if INCLUDE_RTM_OPT
 734   if (UseRTMLocking) {
 735     if (is_intel_family_core()) {
 736       if ((_model == CPU_MODEL_HASWELL_E3) ||
 737           (_model == CPU_MODEL_HASWELL_E7 && _stepping < 3) ||
 738           (_model == CPU_MODEL_BROADWELL  && _stepping < 4)) {
 739         // currently a collision between SKL and HSW_E3
 740         if (!UnlockExperimentalVMOptions && UseAVX < 3) {
 741           vm_exit_during_initialization("UseRTMLocking is only available as experimental option on this platform. It must be enabled via -XX:+UnlockExperimentalVMOptions flag.");
 742         } else {
 743           warning("UseRTMLocking is only available as experimental option on this platform.");
 744         }
 745       }
 746     }
 747     if (!FLAG_IS_CMDLINE(UseRTMLocking)) {
 748       // RTM locking should be used only for applications with
 749       // high lock contention. For now we do not use it by default.
 750       vm_exit_during_initialization("UseRTMLocking flag should be only set on command line");
 751     }
 752     if (!is_power_of_2(RTMTotalCountIncrRate)) {
 753       warning("RTMTotalCountIncrRate must be a power of 2, resetting it to 64");
 754       FLAG_SET_DEFAULT(RTMTotalCountIncrRate, 64);
 755     }
 756     if (RTMAbortRatio < 0 || RTMAbortRatio > 100) {
 757       warning("RTMAbortRatio must be in the range 0 to 100, resetting it to 50");
 758       FLAG_SET_DEFAULT(RTMAbortRatio, 50);
 759     }
 760   } else { // !UseRTMLocking
 761     if (UseRTMForStackLocks) {
 762       if (!FLAG_IS_DEFAULT(UseRTMForStackLocks)) {
 763         warning("UseRTMForStackLocks flag should be off when UseRTMLocking flag is off");
 764       }
 765       FLAG_SET_DEFAULT(UseRTMForStackLocks, false);
 766     }
 767     if (UseRTMDeopt) {
 768       FLAG_SET_DEFAULT(UseRTMDeopt, false);
 769     }
 770     if (PrintPreciseRTMLockingStatistics) {
 771       FLAG_SET_DEFAULT(PrintPreciseRTMLockingStatistics, false);
 772     }
 773   }
 774 #else
 775   if (UseRTMLocking) {
 776     // Only C2 does RTM locking optimization.
 777     // Can't continue because UseRTMLocking affects UseBiasedLocking flag
 778     // setting during arguments processing. See use_biased_locking().
 779     vm_exit_during_initialization("RTM locking optimization is not supported in this VM");
 780   }
 781 #endif
 782 
 783 #ifdef COMPILER2
 784   if (UseFPUForSpilling) {
 785     if (UseSSE < 2) {
 786       // Only supported with SSE2+
 787       FLAG_SET_DEFAULT(UseFPUForSpilling, false);
 788     }
 789   }
 790 #endif
 791 #if defined(COMPILER2) || INCLUDE_JVMCI
 792   if (MaxVectorSize > 0) {
 793     if (!is_power_of_2(MaxVectorSize)) {
 794       warning("MaxVectorSize must be a power of 2");
 795       FLAG_SET_DEFAULT(MaxVectorSize, 64);
 796     }
 797     if (MaxVectorSize > 64) {
 798       FLAG_SET_DEFAULT(MaxVectorSize, 64);
 799     }
 800     if (MaxVectorSize > 16 && (UseAVX == 0 || !os_supports_avx_vectors())) {
 801       // 32 bytes vectors (in YMM) are only supported with AVX+
 802       FLAG_SET_DEFAULT(MaxVectorSize, 16);
 803     }
 804     if (UseSSE < 2) {
 805       // Vectors (in XMM) are only supported with SSE2+
 806       FLAG_SET_DEFAULT(MaxVectorSize, 0);
 807     }
 808 #if defined(COMPILER2) && defined(ASSERT)
 809     if (supports_avx() && PrintMiscellaneous && Verbose && TraceNewVectors) {
 810       tty->print_cr("State of YMM registers after signal handle:");
 811       int nreg = 2 LP64_ONLY(+2);
 812       const char* ymm_name[4] = {"0", "7", "8", "15"};
 813       for (int i = 0; i < nreg; i++) {
 814         tty->print("YMM%s:", ymm_name[i]);
 815         for (int j = 7; j >=0; j--) {
 816           tty->print(" %x", _cpuid_info.ymm_save[i*8 + j]);
 817         }
 818         tty->cr();
 819       }
 820     }
 821 #endif // COMPILER2 && ASSERT
 822   }
 823 #endif // COMPILER2 || INCLUDE_JVMCI
 824 
 825 #ifdef COMPILER2
 826 #ifdef _LP64
 827   if (FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) {
 828     UseMultiplyToLenIntrinsic = true;
 829   }
 830   if (FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) {
 831     UseSquareToLenIntrinsic = true;
 832   }
 833   if (FLAG_IS_DEFAULT(UseMulAddIntrinsic)) {
 834     UseMulAddIntrinsic = true;
 835   }
 836   if (FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) {
 837     UseMontgomeryMultiplyIntrinsic = true;
 838   }
 839   if (FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) {
 840     UseMontgomerySquareIntrinsic = true;
 841   }
 842 #else
 843   if (UseMultiplyToLenIntrinsic) {
 844     if (!FLAG_IS_DEFAULT(UseMultiplyToLenIntrinsic)) {
 845       warning("multiplyToLen intrinsic is not available in 32-bit VM");
 846     }
 847     FLAG_SET_DEFAULT(UseMultiplyToLenIntrinsic, false);
 848   }
 849   if (UseMontgomeryMultiplyIntrinsic) {
 850     if (!FLAG_IS_DEFAULT(UseMontgomeryMultiplyIntrinsic)) {
 851       warning("montgomeryMultiply intrinsic is not available in 32-bit VM");
 852     }
 853     FLAG_SET_DEFAULT(UseMontgomeryMultiplyIntrinsic, false);
 854   }
 855   if (UseMontgomerySquareIntrinsic) {
 856     if (!FLAG_IS_DEFAULT(UseMontgomerySquareIntrinsic)) {
 857       warning("montgomerySquare intrinsic is not available in 32-bit VM");
 858     }
 859     FLAG_SET_DEFAULT(UseMontgomerySquareIntrinsic, false);
 860   }
 861   if (UseSquareToLenIntrinsic) {
 862     if (!FLAG_IS_DEFAULT(UseSquareToLenIntrinsic)) {
 863       warning("squareToLen intrinsic is not available in 32-bit VM");
 864     }
 865     FLAG_SET_DEFAULT(UseSquareToLenIntrinsic, false);
 866   }
 867   if (UseMulAddIntrinsic) {
 868     if (!FLAG_IS_DEFAULT(UseMulAddIntrinsic)) {
 869       warning("mulAdd intrinsic is not available in 32-bit VM");
 870     }
 871     FLAG_SET_DEFAULT(UseMulAddIntrinsic, false);
 872   }
 873 #endif
 874 #endif // COMPILER2
 875 
 876   // On new cpus instructions which update whole XMM register should be used
 877   // to prevent partial register stall due to dependencies on high half.
 878   //
 879   // UseXmmLoadAndClearUpper == true  --> movsd(xmm, mem)
 880   // UseXmmLoadAndClearUpper == false --> movlpd(xmm, mem)
 881   // UseXmmRegToRegMoveAll == true  --> movaps(xmm, xmm), movapd(xmm, xmm).
 882   // UseXmmRegToRegMoveAll == false --> movss(xmm, xmm),  movsd(xmm, xmm).
 883 
 884   if( is_amd() ) { // AMD cpus specific settings
 885     if( supports_sse2() && FLAG_IS_DEFAULT(UseAddressNop) ) {
 886       // Use it on new AMD cpus starting from Opteron.
 887       UseAddressNop = true;
 888     }
 889     if( supports_sse2() && FLAG_IS_DEFAULT(UseNewLongLShift) ) {
 890       // Use it on new AMD cpus starting from Opteron.
 891       UseNewLongLShift = true;
 892     }
 893     if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) {
 894       if (supports_sse4a()) {
 895         UseXmmLoadAndClearUpper = true; // use movsd only on '10h' Opteron
 896       } else {
 897         UseXmmLoadAndClearUpper = false;
 898       }
 899     }
 900     if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) {
 901       if( supports_sse4a() ) {
 902         UseXmmRegToRegMoveAll = true; // use movaps, movapd only on '10h'
 903       } else {
 904         UseXmmRegToRegMoveAll = false;
 905       }
 906     }
 907     if( FLAG_IS_DEFAULT(UseXmmI2F) ) {
 908       if( supports_sse4a() ) {
 909         UseXmmI2F = true;
 910       } else {
 911         UseXmmI2F = false;
 912       }
 913     }
 914     if( FLAG_IS_DEFAULT(UseXmmI2D) ) {
 915       if( supports_sse4a() ) {
 916         UseXmmI2D = true;
 917       } else {
 918         UseXmmI2D = false;
 919       }
 920     }
 921     if( FLAG_IS_DEFAULT(UseSSE42Intrinsics) ) {
 922       if( supports_sse4_2() && UseSSE >= 4 ) {
 923         UseSSE42Intrinsics = true;
 924       }
 925     }
 926 
 927     // some defaults for AMD family 15h
 928     if ( cpu_family() == 0x15 ) {
 929       // On family 15h processors default is no sw prefetch
 930       if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
 931         AllocatePrefetchStyle = 0;
 932       }
 933       // Also, if some other prefetch style is specified, default instruction type is PREFETCHW
 934       if (FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
 935         AllocatePrefetchInstr = 3;
 936       }
 937       // On family 15h processors use XMM and UnalignedLoadStores for Array Copy
 938       if (supports_sse2() && FLAG_IS_DEFAULT(UseXMMForArrayCopy)) {
 939         UseXMMForArrayCopy = true;
 940       }
 941       if (supports_sse2() && FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
 942         UseUnalignedLoadStores = true;
 943       }
 944     }
 945 
 946 #ifdef COMPILER2
 947     if (MaxVectorSize > 16) {
 948       // Limit vectors size to 16 bytes on current AMD cpus.
 949       FLAG_SET_DEFAULT(MaxVectorSize, 16);
 950     }
 951 #endif // COMPILER2
 952   }
 953 
 954   if( is_intel() ) { // Intel cpus specific settings
 955     if( FLAG_IS_DEFAULT(UseStoreImmI16) ) {
 956       UseStoreImmI16 = false; // don't use it on Intel cpus
 957     }
 958     if( cpu_family() == 6 || cpu_family() == 15 ) {
 959       if( FLAG_IS_DEFAULT(UseAddressNop) ) {
 960         // Use it on all Intel cpus starting from PentiumPro
 961         UseAddressNop = true;
 962       }
 963     }
 964     if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) {
 965       UseXmmLoadAndClearUpper = true; // use movsd on all Intel cpus
 966     }
 967     if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) {
 968       if( supports_sse3() ) {
 969         UseXmmRegToRegMoveAll = true; // use movaps, movapd on new Intel cpus
 970       } else {
 971         UseXmmRegToRegMoveAll = false;
 972       }
 973     }
 974     if( cpu_family() == 6 && supports_sse3() ) { // New Intel cpus
 975 #ifdef COMPILER2
 976       if( FLAG_IS_DEFAULT(MaxLoopPad) ) {
 977         // For new Intel cpus do the next optimization:
 978         // don't align the beginning of a loop if there are enough instructions
 979         // left (NumberOfLoopInstrToAlign defined in c2_globals.hpp)
 980         // in current fetch line (OptoLoopAlignment) or the padding
 981         // is big (> MaxLoopPad).
 982         // Set MaxLoopPad to 11 for new Intel cpus to reduce number of
 983         // generated NOP instructions. 11 is the largest size of one
 984         // address NOP instruction '0F 1F' (see Assembler::nop(i)).
 985         MaxLoopPad = 11;
 986       }
 987 #endif // COMPILER2
 988       if (FLAG_IS_DEFAULT(UseXMMForArrayCopy)) {
 989         UseXMMForArrayCopy = true; // use SSE2 movq on new Intel cpus
 990       }
 991       if (supports_sse4_2() && supports_ht()) { // Newest Intel cpus
 992         if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
 993           UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus
 994         }
 995       }
 996       if (supports_sse4_2() && UseSSE >= 4) {
 997         if (FLAG_IS_DEFAULT(UseSSE42Intrinsics)) {
 998           UseSSE42Intrinsics = true;
 999         }
1000       }
1001     }
1002     if ((cpu_family() == 0x06) &&
1003         ((extended_cpu_model() == 0x36) || // Centerton
1004          (extended_cpu_model() == 0x37) || // Silvermont
1005          (extended_cpu_model() == 0x4D))) {
1006 #ifdef COMPILER2
1007       if (FLAG_IS_DEFAULT(OptoScheduling)) {
1008         OptoScheduling = true;
1009       }
1010 #endif
1011       if (supports_sse4_2()) { // Silvermont
1012         if (FLAG_IS_DEFAULT(UseUnalignedLoadStores)) {
1013           UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus
1014         }
1015       }
1016     }
1017     if(FLAG_IS_DEFAULT(AllocatePrefetchInstr) && supports_3dnow_prefetch()) {
1018       AllocatePrefetchInstr = 3;
1019     }
1020   }
1021 
1022   // Use count leading zeros count instruction if available.
1023   if (supports_lzcnt()) {
1024     if (FLAG_IS_DEFAULT(UseCountLeadingZerosInstruction)) {
1025       UseCountLeadingZerosInstruction = true;
1026     }
1027    } else if (UseCountLeadingZerosInstruction) {
1028     warning("lzcnt instruction is not available on this CPU");
1029     FLAG_SET_DEFAULT(UseCountLeadingZerosInstruction, false);
1030   }
1031 
1032   // Use count trailing zeros instruction if available
1033   if (supports_bmi1()) {
1034     // tzcnt does not require VEX prefix
1035     if (FLAG_IS_DEFAULT(UseCountTrailingZerosInstruction)) {
1036       if (!UseBMI1Instructions && !FLAG_IS_DEFAULT(UseBMI1Instructions)) {
1037         // Don't use tzcnt if BMI1 is switched off on command line.
1038         UseCountTrailingZerosInstruction = false;
1039       } else {
1040         UseCountTrailingZerosInstruction = true;
1041       }
1042     }
1043   } else if (UseCountTrailingZerosInstruction) {
1044     warning("tzcnt instruction is not available on this CPU");
1045     FLAG_SET_DEFAULT(UseCountTrailingZerosInstruction, false);
1046   }
1047 
1048   // BMI instructions (except tzcnt) use an encoding with VEX prefix.
1049   // VEX prefix is generated only when AVX > 0.
1050   if (supports_bmi1() && supports_avx()) {
1051     if (FLAG_IS_DEFAULT(UseBMI1Instructions)) {
1052       UseBMI1Instructions = true;
1053     }
1054   } else if (UseBMI1Instructions) {
1055     warning("BMI1 instructions are not available on this CPU (AVX is also required)");
1056     FLAG_SET_DEFAULT(UseBMI1Instructions, false);
1057   }
1058 
1059   if (supports_bmi2() && supports_avx()) {
1060     if (FLAG_IS_DEFAULT(UseBMI2Instructions)) {
1061       UseBMI2Instructions = true;
1062     }
1063   } else if (UseBMI2Instructions) {
1064     warning("BMI2 instructions are not available on this CPU (AVX is also required)");
1065     FLAG_SET_DEFAULT(UseBMI2Instructions, false);
1066   }
1067 
1068   // Use population count instruction if available.
1069   if (supports_popcnt()) {
1070     if (FLAG_IS_DEFAULT(UsePopCountInstruction)) {
1071       UsePopCountInstruction = true;
1072     }
1073   } else if (UsePopCountInstruction) {
1074     warning("POPCNT instruction is not available on this CPU");
1075     FLAG_SET_DEFAULT(UsePopCountInstruction, false);
1076   }
1077 
1078   // Use fast-string operations if available.
1079   if (supports_erms()) {
1080     if (FLAG_IS_DEFAULT(UseFastStosb)) {
1081       UseFastStosb = true;
1082     }
1083   } else if (UseFastStosb) {
1084     warning("fast-string operations are not available on this CPU");
1085     FLAG_SET_DEFAULT(UseFastStosb, false);
1086   }
1087 
1088 #ifdef COMPILER2
1089   if (FLAG_IS_DEFAULT(AlignVector)) {
1090     // Modern processors allow misaligned memory operations for vectors.
1091     AlignVector = !UseUnalignedLoadStores;
1092   }
1093 #endif // COMPILER2
1094 
1095   if( AllocatePrefetchInstr == 3 && !supports_3dnow_prefetch() ) AllocatePrefetchInstr=0;
1096   if( !supports_sse() && supports_3dnow_prefetch() ) AllocatePrefetchInstr = 3;
1097 
1098   // Allocation prefetch settings
1099   intx cache_line_size = prefetch_data_size();
1100   if( cache_line_size > AllocatePrefetchStepSize )
1101     AllocatePrefetchStepSize = cache_line_size;
1102 
1103   assert(AllocatePrefetchLines > 0, "invalid value");
1104   if( AllocatePrefetchLines < 1 )     // set valid value in product VM
1105     AllocatePrefetchLines = 3;
1106   assert(AllocateInstancePrefetchLines > 0, "invalid value");
1107   if( AllocateInstancePrefetchLines < 1 ) // set valid value in product VM
1108     AllocateInstancePrefetchLines = 1;
1109 
1110   AllocatePrefetchDistance = allocate_prefetch_distance();
1111   AllocatePrefetchStyle    = allocate_prefetch_style();
1112 
1113   if (is_intel() && cpu_family() == 6 && supports_sse3()) {
1114     if (AllocatePrefetchStyle == 2) { // watermark prefetching on Core
1115 #ifdef _LP64
1116       AllocatePrefetchDistance = 384;
1117 #else
1118       AllocatePrefetchDistance = 320;
1119 #endif
1120     }
1121     if (supports_sse4_2() && supports_ht()) { // Nehalem based cpus
1122       AllocatePrefetchDistance = 192;
1123       AllocatePrefetchLines = 4;
1124     }
1125 #ifdef COMPILER2
1126     if (supports_sse4_2()) {
1127       if (FLAG_IS_DEFAULT(UseFPUForSpilling)) {
1128         FLAG_SET_DEFAULT(UseFPUForSpilling, true);
1129       }
1130     }
1131 #endif
1132   }
1133 
1134 #ifdef _LP64
1135   // Prefetch settings
1136   PrefetchCopyIntervalInBytes = prefetch_copy_interval_in_bytes();
1137   PrefetchScanIntervalInBytes = prefetch_scan_interval_in_bytes();
1138   PrefetchFieldsAhead         = prefetch_fields_ahead();
1139 #endif
1140 
1141   if (FLAG_IS_DEFAULT(ContendedPaddingWidth) &&
1142      (cache_line_size > ContendedPaddingWidth))
1143      ContendedPaddingWidth = cache_line_size;
1144 
1145   // This machine allows unaligned memory accesses
1146   if (FLAG_IS_DEFAULT(UseUnalignedAccesses)) {
1147     FLAG_SET_DEFAULT(UseUnalignedAccesses, true);
1148   }
1149 
1150 #ifndef PRODUCT
1151   if (PrintMiscellaneous && Verbose) {
1152     tty->print_cr("Logical CPUs per core: %u",
1153                   logical_processors_per_package());
1154     tty->print_cr("L1 data cache line size: %u", L1_data_cache_line_size());
1155     tty->print("UseSSE=%d", (int) UseSSE);
1156     if (UseAVX > 0) {
1157       tty->print("  UseAVX=%d", (int) UseAVX);
1158     }
1159     if (UseAES) {
1160       tty->print("  UseAES=1");
1161     }
1162 #ifdef COMPILER2
1163     if (MaxVectorSize > 0) {
1164       tty->print("  MaxVectorSize=%d", (int) MaxVectorSize);
1165     }
1166 #endif
1167     tty->cr();
1168     tty->print("Allocation");
1169     if (AllocatePrefetchStyle <= 0 || UseSSE == 0 && !supports_3dnow_prefetch()) {
1170       tty->print_cr(": no prefetching");
1171     } else {
1172       tty->print(" prefetching: ");
1173       if (UseSSE == 0 && supports_3dnow_prefetch()) {
1174         tty->print("PREFETCHW");
1175       } else if (UseSSE >= 1) {
1176         if (AllocatePrefetchInstr == 0) {
1177           tty->print("PREFETCHNTA");
1178         } else if (AllocatePrefetchInstr == 1) {
1179           tty->print("PREFETCHT0");
1180         } else if (AllocatePrefetchInstr == 2) {
1181           tty->print("PREFETCHT2");
1182         } else if (AllocatePrefetchInstr == 3) {
1183           tty->print("PREFETCHW");
1184         }
1185       }
1186       if (AllocatePrefetchLines > 1) {
1187         tty->print_cr(" at distance %d, %d lines of %d bytes", (int) AllocatePrefetchDistance, (int) AllocatePrefetchLines, (int) AllocatePrefetchStepSize);
1188       } else {
1189         tty->print_cr(" at distance %d, one line of %d bytes", (int) AllocatePrefetchDistance, (int) AllocatePrefetchStepSize);
1190       }
1191     }
1192 
1193     if (PrefetchCopyIntervalInBytes > 0) {
1194       tty->print_cr("PrefetchCopyIntervalInBytes %d", (int) PrefetchCopyIntervalInBytes);
1195     }
1196     if (PrefetchScanIntervalInBytes > 0) {
1197       tty->print_cr("PrefetchScanIntervalInBytes %d", (int) PrefetchScanIntervalInBytes);
1198     }
1199     if (PrefetchFieldsAhead > 0) {
1200       tty->print_cr("PrefetchFieldsAhead %d", (int) PrefetchFieldsAhead);
1201     }
1202     if (ContendedPaddingWidth > 0) {
1203       tty->print_cr("ContendedPaddingWidth %d", (int) ContendedPaddingWidth);
1204     }
1205   }
1206 #endif // !PRODUCT
1207 }
1208 
1209 bool VM_Version::use_biased_locking() {
1210 #if INCLUDE_RTM_OPT
1211   // RTM locking is most useful when there is high lock contention and
1212   // low data contention.  With high lock contention the lock is usually
1213   // inflated and biased locking is not suitable for that case.
1214   // RTM locking code requires that biased locking is off.
1215   // Note: we can't switch off UseBiasedLocking in get_processor_features()
1216   // because it is used by Thread::allocate() which is called before
1217   // VM_Version::initialize().
1218   if (UseRTMLocking && UseBiasedLocking) {
1219     if (FLAG_IS_DEFAULT(UseBiasedLocking)) {
1220       FLAG_SET_DEFAULT(UseBiasedLocking, false);
1221     } else {
1222       warning("Biased locking is not supported with RTM locking; ignoring UseBiasedLocking flag." );
1223       UseBiasedLocking = false;
1224     }
1225   }
1226 #endif
1227   return UseBiasedLocking;
1228 }
1229 
1230 void VM_Version::initialize() {
1231   ResourceMark rm;
1232   // Making this stub must be FIRST use of assembler
1233 
1234   stub_blob = BufferBlob::create("get_cpu_info_stub", stub_size);
1235   if (stub_blob == NULL) {
1236     vm_exit_during_initialization("Unable to allocate get_cpu_info_stub");
1237   }
1238   CodeBuffer c(stub_blob);
1239   VM_Version_StubGenerator g(&c);
1240   get_cpu_info_stub = CAST_TO_FN_PTR(get_cpu_info_stub_t,
1241                                      g.generate_get_cpu_info());
1242 
1243   get_processor_features();
1244 }