478 result |= _cpuid_info.std_cpuid1_eax.bits.ext_model << 4; 479 return result; 480 } 481 482 static uint32_t cpu_stepping() { 483 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.stepping; 484 return result; 485 } 486 487 static uint logical_processor_count() { 488 uint result = threads_per_core(); 489 return result; 490 } 491 492 static uint64_t feature_flags() { 493 uint64_t result = 0; 494 if (_cpuid_info.std_cpuid1_edx.bits.cmpxchg8 != 0) 495 result |= CPU_CX8; 496 if (_cpuid_info.std_cpuid1_edx.bits.cmov != 0) 497 result |= CPU_CMOV; 498 if (_cpuid_info.std_cpuid1_edx.bits.fxsr != 0 || (is_amd() && 499 _cpuid_info.ext_cpuid1_edx.bits.fxsr != 0)) 500 result |= CPU_FXSR; 501 // HT flag is set for multi-core processors also. 502 if (threads_per_core() > 1) 503 result |= CPU_HT; 504 if (_cpuid_info.std_cpuid1_edx.bits.mmx != 0 || (is_amd() && 505 _cpuid_info.ext_cpuid1_edx.bits.mmx != 0)) 506 result |= CPU_MMX; 507 if (_cpuid_info.std_cpuid1_edx.bits.sse != 0) 508 result |= CPU_SSE; 509 if (_cpuid_info.std_cpuid1_edx.bits.sse2 != 0) 510 result |= CPU_SSE2; 511 if (_cpuid_info.std_cpuid1_ecx.bits.sse3 != 0) 512 result |= CPU_SSE3; 513 if (_cpuid_info.std_cpuid1_ecx.bits.ssse3 != 0) 514 result |= CPU_SSSE3; 515 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_1 != 0) 516 result |= CPU_SSE4_1; 517 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_2 != 0) 518 result |= CPU_SSE4_2; 519 if (_cpuid_info.std_cpuid1_ecx.bits.popcnt != 0) 520 result |= CPU_POPCNT; 521 if (_cpuid_info.std_cpuid1_ecx.bits.avx != 0 && 522 _cpuid_info.std_cpuid1_ecx.bits.osxsave != 0 && 523 _cpuid_info.xem_xcr0_eax.bits.sse != 0 && 524 _cpuid_info.xem_xcr0_eax.bits.ymm != 0) { 559 result |= CPU_TSC; 560 if (_cpuid_info.ext_cpuid7_edx.bits.tsc_invariance != 0) 561 result |= CPU_TSCINV; 562 if (_cpuid_info.std_cpuid1_ecx.bits.aes != 0) 563 result |= CPU_AES; 564 if (_cpuid_info.sef_cpuid7_ebx.bits.erms != 0) 565 result |= CPU_ERMS; 566 if (_cpuid_info.std_cpuid1_ecx.bits.clmul != 0) 567 result |= CPU_CLMUL; 568 if (_cpuid_info.sef_cpuid7_ebx.bits.rtm != 0) 569 result |= CPU_RTM; 570 if(_cpuid_info.sef_cpuid7_ebx.bits.adx != 0) 571 result |= CPU_ADX; 572 if(_cpuid_info.sef_cpuid7_ebx.bits.bmi2 != 0) 573 result |= CPU_BMI2; 574 if (_cpuid_info.sef_cpuid7_ebx.bits.sha != 0) 575 result |= CPU_SHA; 576 if (_cpuid_info.std_cpuid1_ecx.bits.fma != 0) 577 result |= CPU_FMA; 578 579 // AMD features. 580 if (is_amd()) { 581 if ((_cpuid_info.ext_cpuid1_edx.bits.tdnow != 0) || 582 (_cpuid_info.ext_cpuid1_ecx.bits.prefetchw != 0)) 583 result |= CPU_3DNOW_PREFETCH; 584 if (_cpuid_info.ext_cpuid1_ecx.bits.lzcnt != 0) 585 result |= CPU_LZCNT; 586 if (_cpuid_info.ext_cpuid1_ecx.bits.sse4a != 0) 587 result |= CPU_SSE4A; 588 } 589 // Intel features. 590 if(is_intel()) { 591 if(_cpuid_info.ext_cpuid1_ecx.bits.lzcnt_intel != 0) 592 result |= CPU_LZCNT; 593 // for Intel, ecx.bits.misalignsse bit (bit 8) indicates support for prefetchw 594 if (_cpuid_info.ext_cpuid1_ecx.bits.misalignsse != 0) { 595 result |= CPU_3DNOW_PREFETCH; 596 } 597 } 598 599 // ZX features. 600 if (is_zx()) { 694 } 695 696 // 697 // Processor family: 698 // 3 - 386 699 // 4 - 486 700 // 5 - Pentium 701 // 6 - PentiumPro, Pentium II, Celeron, Xeon, Pentium III, Athlon, 702 // Pentium M, Core Solo, Core Duo, Core2 Duo 703 // family 6 model: 9, 13, 14, 15 704 // 0x0f - Pentium 4, Opteron 705 // 706 // Note: The cpu family should be used to select between 707 // instruction sequences which are valid on all Intel 708 // processors. Use the feature test functions below to 709 // determine whether a particular instruction is supported. 710 // 711 static int cpu_family() { return _cpu;} 712 static bool is_P6() { return cpu_family() >= 6; } 713 static bool is_amd() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x68747541; } // 'htuA' 714 static bool is_intel() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x756e6547; } // 'uneG' 715 static bool is_zx() { assert_is_initialized(); return (_cpuid_info.std_vendor_name_0 == 0x746e6543) || (_cpuid_info.std_vendor_name_0 == 0x68532020); } // 'tneC'||'hS ' 716 static bool is_atom_family() { return ((cpu_family() == 0x06) && ((extended_cpu_model() == 0x36) || (extended_cpu_model() == 0x37) || (extended_cpu_model() == 0x4D))); } //Silvermont and Centerton 717 static bool is_knights_family() { return ((cpu_family() == 0x06) && ((extended_cpu_model() == 0x57) || (extended_cpu_model() == 0x85))); } // Xeon Phi 3200/5200/7200 and Future Xeon Phi 718 719 static bool supports_processor_topology() { 720 return (_cpuid_info.std_max_function >= 0xB) && 721 // eax[4:0] | ebx[0:15] == 0 indicates invalid topology level. 722 // Some cpus have max cpuid >= 0xB but do not support processor topology. 723 (((_cpuid_info.tpl_cpuidB0_eax & 0x1f) | _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus) != 0); 724 } 725 726 static uint cores_per_cpu() { 727 uint result = 1; 728 if (is_intel()) { 729 bool supports_topology = supports_processor_topology(); 730 if (supports_topology) { 731 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus / 732 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 733 } 734 if (!supports_topology || result == 0) { 735 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1); 736 } 737 } else if (is_amd()) { 738 result = (_cpuid_info.ext_cpuid8_ecx.bits.cores_per_cpu + 1); 739 } else if (is_zx()) { 740 bool supports_topology = supports_processor_topology(); 741 if (supports_topology) { 742 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus / 743 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 744 } 745 if (!supports_topology || result == 0) { 746 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1); 747 } 748 } 749 return result; 750 } 751 752 static uint threads_per_core() { 753 uint result = 1; 754 if (is_intel() && supports_processor_topology()) { 755 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 756 } else if (is_zx() && supports_processor_topology()) { 757 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 758 } else if (_cpuid_info.std_cpuid1_edx.bits.ht != 0) { 759 if (cpu_family() >= 0x17) { 760 result = _cpuid_info.ext_cpuid1E_ebx.bits.threads_per_core + 1; 761 } else { 762 result = _cpuid_info.std_cpuid1_ebx.bits.threads_per_cpu / 763 cores_per_cpu(); 764 } 765 } 766 return (result == 0 ? 1 : result); 767 } 768 769 static intx L1_line_size() { 770 intx result = 0; 771 if (is_intel()) { 772 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1); 773 } else if (is_amd()) { 774 result = _cpuid_info.ext_cpuid5_ecx.bits.L1_line_size; 775 } else if (is_zx()) { 776 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1); 777 } 778 if (result < 32) // not defined ? 779 result = 32; // 32 bytes by default on x86 and other x64 780 return result; 781 } 782 783 static intx prefetch_data_size() { 784 return L1_line_size(); 785 } 786 787 // 788 // Feature identification 789 // 790 static bool supports_cpuid() { return _features != 0; } 791 static bool supports_cmpxchg8() { return (_features & CPU_CX8) != 0; } 792 static bool supports_cmov() { return (_features & CPU_CMOV) != 0; } 793 static bool supports_fxsr() { return (_features & CPU_FXSR) != 0; } 840 static bool is_intel_tsc_synched_at_init() { 841 if (is_intel_family_core()) { 842 uint32_t ext_model = extended_cpu_model(); 843 if (ext_model == CPU_MODEL_NEHALEM_EP || 844 ext_model == CPU_MODEL_WESTMERE_EP || 845 ext_model == CPU_MODEL_SANDYBRIDGE_EP || 846 ext_model == CPU_MODEL_IVYBRIDGE_EP) { 847 // <= 2-socket invariant tsc support. EX versions are usually used 848 // in > 2-socket systems and likely don't synchronize tscs at 849 // initialization. 850 // Code that uses tsc values must be prepared for them to arbitrarily 851 // jump forward or backward. 852 return true; 853 } 854 } 855 return false; 856 } 857 858 // AMD features 859 static bool supports_3dnow_prefetch() { return (_features & CPU_3DNOW_PREFETCH) != 0; } 860 static bool supports_mmx_ext() { return is_amd() && _cpuid_info.ext_cpuid1_edx.bits.mmx_amd != 0; } 861 static bool supports_lzcnt() { return (_features & CPU_LZCNT) != 0; } 862 static bool supports_sse4a() { return (_features & CPU_SSE4A) != 0; } 863 864 static bool is_amd_Barcelona() { return is_amd() && 865 extended_cpu_family() == CPU_FAMILY_AMD_11H; } 866 867 // Intel and AMD newer cores support fast timestamps well 868 static bool supports_tscinv_bit() { 869 return (_features & CPU_TSCINV) != 0; 870 } 871 static bool supports_tscinv() { 872 return supports_tscinv_bit() && 873 ( (is_amd() && !is_amd_Barcelona()) || 874 is_intel_tsc_synched_at_init() ); 875 } 876 877 // Intel Core and newer cpus have fast IDIV instruction (excluding Atom). 878 static bool has_fast_idiv() { return is_intel() && cpu_family() == 6 && 879 supports_sse3() && _model != 0x1C; } 880 881 static bool supports_compare_and_exchange() { return true; } 882 883 static intx allocate_prefetch_distance(bool use_watermark_prefetch) { 884 // Hardware prefetching (distance/size in bytes): 885 // Pentium 3 - 64 / 32 886 // Pentium 4 - 256 / 128 887 // Athlon - 64 / 32 ???? 888 // Opteron - 128 / 64 only when 2 sequential cache lines accessed 889 // Core - 128 / 64 890 // 891 // Software prefetching (distance in bytes / instruction with best score): 892 // Pentium 3 - 128 / prefetchnta 893 // Pentium 4 - 512 / prefetchnta 894 // Athlon - 128 / prefetchnta 895 // Opteron - 256 / prefetchnta 896 // Core - 256 / prefetchnta 897 // It will be used only when AllocatePrefetchStyle > 0 898 899 if (is_amd()) { // AMD 900 if (supports_sse2()) { 901 return 256; // Opteron 902 } else { 903 return 128; // Athlon 904 } 905 } else { // Intel 906 if (supports_sse3() && cpu_family() == 6) { 907 if (supports_sse4_2() && supports_ht()) { // Nehalem based cpus 908 return 192; 909 } else if (use_watermark_prefetch) { // watermark prefetching on Core 910 #ifdef _LP64 911 return 384; 912 #else 913 return 320; 914 #endif 915 } 916 } 917 if (supports_sse2()) { 918 if (cpu_family() == 6) { 919 return 256; // Pentium M, Core, Core2 | 478 result |= _cpuid_info.std_cpuid1_eax.bits.ext_model << 4; 479 return result; 480 } 481 482 static uint32_t cpu_stepping() { 483 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.stepping; 484 return result; 485 } 486 487 static uint logical_processor_count() { 488 uint result = threads_per_core(); 489 return result; 490 } 491 492 static uint64_t feature_flags() { 493 uint64_t result = 0; 494 if (_cpuid_info.std_cpuid1_edx.bits.cmpxchg8 != 0) 495 result |= CPU_CX8; 496 if (_cpuid_info.std_cpuid1_edx.bits.cmov != 0) 497 result |= CPU_CMOV; 498 if (_cpuid_info.std_cpuid1_edx.bits.fxsr != 0 || ((is_amd() || is_hygon()) && 499 _cpuid_info.ext_cpuid1_edx.bits.fxsr != 0)) 500 result |= CPU_FXSR; 501 // HT flag is set for multi-core processors also. 502 if (threads_per_core() > 1) 503 result |= CPU_HT; 504 if (_cpuid_info.std_cpuid1_edx.bits.mmx != 0 || ((is_amd() || is_hygon()) && 505 _cpuid_info.ext_cpuid1_edx.bits.mmx != 0)) 506 result |= CPU_MMX; 507 if (_cpuid_info.std_cpuid1_edx.bits.sse != 0) 508 result |= CPU_SSE; 509 if (_cpuid_info.std_cpuid1_edx.bits.sse2 != 0) 510 result |= CPU_SSE2; 511 if (_cpuid_info.std_cpuid1_ecx.bits.sse3 != 0) 512 result |= CPU_SSE3; 513 if (_cpuid_info.std_cpuid1_ecx.bits.ssse3 != 0) 514 result |= CPU_SSSE3; 515 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_1 != 0) 516 result |= CPU_SSE4_1; 517 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_2 != 0) 518 result |= CPU_SSE4_2; 519 if (_cpuid_info.std_cpuid1_ecx.bits.popcnt != 0) 520 result |= CPU_POPCNT; 521 if (_cpuid_info.std_cpuid1_ecx.bits.avx != 0 && 522 _cpuid_info.std_cpuid1_ecx.bits.osxsave != 0 && 523 _cpuid_info.xem_xcr0_eax.bits.sse != 0 && 524 _cpuid_info.xem_xcr0_eax.bits.ymm != 0) { 559 result |= CPU_TSC; 560 if (_cpuid_info.ext_cpuid7_edx.bits.tsc_invariance != 0) 561 result |= CPU_TSCINV; 562 if (_cpuid_info.std_cpuid1_ecx.bits.aes != 0) 563 result |= CPU_AES; 564 if (_cpuid_info.sef_cpuid7_ebx.bits.erms != 0) 565 result |= CPU_ERMS; 566 if (_cpuid_info.std_cpuid1_ecx.bits.clmul != 0) 567 result |= CPU_CLMUL; 568 if (_cpuid_info.sef_cpuid7_ebx.bits.rtm != 0) 569 result |= CPU_RTM; 570 if(_cpuid_info.sef_cpuid7_ebx.bits.adx != 0) 571 result |= CPU_ADX; 572 if(_cpuid_info.sef_cpuid7_ebx.bits.bmi2 != 0) 573 result |= CPU_BMI2; 574 if (_cpuid_info.sef_cpuid7_ebx.bits.sha != 0) 575 result |= CPU_SHA; 576 if (_cpuid_info.std_cpuid1_ecx.bits.fma != 0) 577 result |= CPU_FMA; 578 579 // AMD|Hygon features. 580 if (is_amd() || is_hygon()) { 581 if ((_cpuid_info.ext_cpuid1_edx.bits.tdnow != 0) || 582 (_cpuid_info.ext_cpuid1_ecx.bits.prefetchw != 0)) 583 result |= CPU_3DNOW_PREFETCH; 584 if (_cpuid_info.ext_cpuid1_ecx.bits.lzcnt != 0) 585 result |= CPU_LZCNT; 586 if (_cpuid_info.ext_cpuid1_ecx.bits.sse4a != 0) 587 result |= CPU_SSE4A; 588 } 589 // Intel features. 590 if(is_intel()) { 591 if(_cpuid_info.ext_cpuid1_ecx.bits.lzcnt_intel != 0) 592 result |= CPU_LZCNT; 593 // for Intel, ecx.bits.misalignsse bit (bit 8) indicates support for prefetchw 594 if (_cpuid_info.ext_cpuid1_ecx.bits.misalignsse != 0) { 595 result |= CPU_3DNOW_PREFETCH; 596 } 597 } 598 599 // ZX features. 600 if (is_zx()) { 694 } 695 696 // 697 // Processor family: 698 // 3 - 386 699 // 4 - 486 700 // 5 - Pentium 701 // 6 - PentiumPro, Pentium II, Celeron, Xeon, Pentium III, Athlon, 702 // Pentium M, Core Solo, Core Duo, Core2 Duo 703 // family 6 model: 9, 13, 14, 15 704 // 0x0f - Pentium 4, Opteron 705 // 706 // Note: The cpu family should be used to select between 707 // instruction sequences which are valid on all Intel 708 // processors. Use the feature test functions below to 709 // determine whether a particular instruction is supported. 710 // 711 static int cpu_family() { return _cpu;} 712 static bool is_P6() { return cpu_family() >= 6; } 713 static bool is_amd() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x68747541; } // 'htuA' 714 static bool is_hygon() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x6F677948; } // 'ogyH' 715 static bool is_intel() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x756e6547; } // 'uneG' 716 static bool is_zx() { assert_is_initialized(); return (_cpuid_info.std_vendor_name_0 == 0x746e6543) || (_cpuid_info.std_vendor_name_0 == 0x68532020); } // 'tneC'||'hS ' 717 static bool is_atom_family() { return ((cpu_family() == 0x06) && ((extended_cpu_model() == 0x36) || (extended_cpu_model() == 0x37) || (extended_cpu_model() == 0x4D))); } //Silvermont and Centerton 718 static bool is_knights_family() { return ((cpu_family() == 0x06) && ((extended_cpu_model() == 0x57) || (extended_cpu_model() == 0x85))); } // Xeon Phi 3200/5200/7200 and Future Xeon Phi 719 720 static bool supports_processor_topology() { 721 return (_cpuid_info.std_max_function >= 0xB) && 722 // eax[4:0] | ebx[0:15] == 0 indicates invalid topology level. 723 // Some cpus have max cpuid >= 0xB but do not support processor topology. 724 (((_cpuid_info.tpl_cpuidB0_eax & 0x1f) | _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus) != 0); 725 } 726 727 static uint cores_per_cpu() { 728 uint result = 1; 729 if (is_intel()) { 730 bool supports_topology = supports_processor_topology(); 731 if (supports_topology) { 732 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus / 733 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 734 } 735 if (!supports_topology || result == 0) { 736 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1); 737 } 738 } else if (is_amd() || is_hygon()) { 739 result = (_cpuid_info.ext_cpuid8_ecx.bits.cores_per_cpu + 1); 740 } else if (is_zx()) { 741 bool supports_topology = supports_processor_topology(); 742 if (supports_topology) { 743 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus / 744 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 745 } 746 if (!supports_topology || result == 0) { 747 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1); 748 } 749 } 750 return result; 751 } 752 753 static uint threads_per_core() { 754 uint result = 1; 755 if (is_intel() && supports_processor_topology()) { 756 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 757 } else if (is_zx() && supports_processor_topology()) { 758 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; 759 } else if (_cpuid_info.std_cpuid1_edx.bits.ht != 0) { 760 if (cpu_family() >= 0x17) { 761 result = _cpuid_info.ext_cpuid1E_ebx.bits.threads_per_core + 1; 762 } else { 763 result = _cpuid_info.std_cpuid1_ebx.bits.threads_per_cpu / 764 cores_per_cpu(); 765 } 766 } 767 return (result == 0 ? 1 : result); 768 } 769 770 static intx L1_line_size() { 771 intx result = 0; 772 if (is_intel()) { 773 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1); 774 } else if (is_amd() || is_hygon()) { 775 result = _cpuid_info.ext_cpuid5_ecx.bits.L1_line_size; 776 } else if (is_zx()) { 777 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1); 778 } 779 if (result < 32) // not defined ? 780 result = 32; // 32 bytes by default on x86 and other x64 781 return result; 782 } 783 784 static intx prefetch_data_size() { 785 return L1_line_size(); 786 } 787 788 // 789 // Feature identification 790 // 791 static bool supports_cpuid() { return _features != 0; } 792 static bool supports_cmpxchg8() { return (_features & CPU_CX8) != 0; } 793 static bool supports_cmov() { return (_features & CPU_CMOV) != 0; } 794 static bool supports_fxsr() { return (_features & CPU_FXSR) != 0; } 841 static bool is_intel_tsc_synched_at_init() { 842 if (is_intel_family_core()) { 843 uint32_t ext_model = extended_cpu_model(); 844 if (ext_model == CPU_MODEL_NEHALEM_EP || 845 ext_model == CPU_MODEL_WESTMERE_EP || 846 ext_model == CPU_MODEL_SANDYBRIDGE_EP || 847 ext_model == CPU_MODEL_IVYBRIDGE_EP) { 848 // <= 2-socket invariant tsc support. EX versions are usually used 849 // in > 2-socket systems and likely don't synchronize tscs at 850 // initialization. 851 // Code that uses tsc values must be prepared for them to arbitrarily 852 // jump forward or backward. 853 return true; 854 } 855 } 856 return false; 857 } 858 859 // AMD features 860 static bool supports_3dnow_prefetch() { return (_features & CPU_3DNOW_PREFETCH) != 0; } 861 static bool supports_mmx_ext() { return (is_amd()||is_hygon()) && _cpuid_info.ext_cpuid1_edx.bits.mmx_amd != 0; } 862 static bool supports_lzcnt() { return (_features & CPU_LZCNT) != 0; } 863 static bool supports_sse4a() { return (_features & CPU_SSE4A) != 0; } 864 865 static bool is_amd_Barcelona() { return is_amd() && 866 extended_cpu_family() == CPU_FAMILY_AMD_11H; } 867 868 // Intel and AMD newer cores support fast timestamps well 869 static bool supports_tscinv_bit() { 870 return (_features & CPU_TSCINV) != 0; 871 } 872 static bool supports_tscinv() { 873 return supports_tscinv_bit() && 874 ( ((is_amd()||is_hygon()) && !is_amd_Barcelona()) || 875 is_intel_tsc_synched_at_init() ); 876 } 877 878 // Intel Core and newer cpus have fast IDIV instruction (excluding Atom). 879 static bool has_fast_idiv() { return is_intel() && cpu_family() == 6 && 880 supports_sse3() && _model != 0x1C; } 881 882 static bool supports_compare_and_exchange() { return true; } 883 884 static intx allocate_prefetch_distance(bool use_watermark_prefetch) { 885 // Hardware prefetching (distance/size in bytes): 886 // Pentium 3 - 64 / 32 887 // Pentium 4 - 256 / 128 888 // Athlon - 64 / 32 ???? 889 // Opteron - 128 / 64 only when 2 sequential cache lines accessed 890 // Core - 128 / 64 891 // 892 // Software prefetching (distance in bytes / instruction with best score): 893 // Pentium 3 - 128 / prefetchnta 894 // Pentium 4 - 512 / prefetchnta 895 // Athlon - 128 / prefetchnta 896 // Opteron - 256 / prefetchnta 897 // Core - 256 / prefetchnta 898 // It will be used only when AllocatePrefetchStyle > 0 899 900 if (is_amd() || is_hygon()) { // AMD 901 if (supports_sse2()) { 902 return 256; // Opteron 903 } else { 904 return 128; // Athlon 905 } 906 } else { // Intel 907 if (supports_sse3() && cpu_family() == 6) { 908 if (supports_sse4_2() && supports_ht()) { // Nehalem based cpus 909 return 192; 910 } else if (use_watermark_prefetch) { // watermark prefetching on Core 911 #ifdef _LP64 912 return 384; 913 #else 914 return 320; 915 #endif 916 } 917 } 918 if (supports_sse2()) { 919 if (cpu_family() == 6) { 920 return 256; // Pentium M, Core, Core2 |