/* SPDX-License-Identifier: BSL-1.0 OR BSD-3-Clause */ #ifndef MPT_ARCH_X86_AMD64_HPP #define MPT_ARCH_X86_AMD64_HPP #include "mpt/arch/feature_flags.hpp" #include "mpt/base/bit.hpp" #include "mpt/base/detect.hpp" #include "mpt/base/integer.hpp" #include "mpt/base/macros.hpp" #include "mpt/base/namespace.hpp" #include "mpt/base/utility.hpp" #include "mpt/osinfo/windows_version.hpp" #include #include #if MPT_ARCH_X86 || MPT_ARCH_AMD64 #if MPT_COMPILER_GCC #include #endif #endif #include #include #include #include #if MPT_OS_DJGPP #include #include #endif #if MPT_OS_WINDOWS #include #endif #if MPT_ARCH_X86 || MPT_ARCH_AMD64 #if MPT_COMPILER_MSVC #include #endif #if MPT_COMPILER_MSVC #include #elif MPT_COMPILER_GCC || MPT_COMPILER_CLANG #include #include #endif #endif // skip assumed features for amd64 #define MPT_ARCH_X86_AMD64_FAST_DETECT 0 namespace mpt { inline namespace MPT_INLINE_NS { namespace arch { namespace x86 { using feature_flags = mpt::arch::basic_feature_flags; using mode_flags = mpt::arch::basic_feature_flags; // clang-format off namespace feature { inline constexpr feature_flags none = feature_flags{}; inline constexpr feature_flags intel386 = feature_flags{ 0x0000'0001 }; inline constexpr feature_flags fpu = feature_flags{ 0x0000'0002 }; inline constexpr feature_flags fsin = feature_flags{ 0x0000'0004 }; inline constexpr feature_flags intel486 = feature_flags{ 0x0000'0008 }; // XADD, BSWAP, CMPXCHG inline constexpr feature_flags cpuid = feature_flags{ 0x0000'0010 }; inline constexpr feature_flags tsc = feature_flags{ 0x0000'0020 }; inline constexpr feature_flags cx8 = feature_flags{ 0x0000'0040 }; inline constexpr feature_flags cmov = feature_flags{ 0x0000'0080 }; inline constexpr feature_flags mmx = feature_flags{ 0x0000'0100 }; inline constexpr feature_flags mmxext = feature_flags{ 0x0000'0200 }; inline constexpr feature_flags x3dnow = feature_flags{ 0x0000'0400 }; inline constexpr feature_flags x3dnowext = feature_flags{ 0x0000'0800 }; inline constexpr feature_flags x3dnowprefetch = feature_flags{ 0x0000'1000 }; inline constexpr feature_flags fxsr = feature_flags{ 0x0000'2000 }; inline constexpr feature_flags sse = feature_flags{ 0x0000'4000 }; inline constexpr feature_flags sse2 = feature_flags{ 0x0000'8000 }; inline constexpr feature_flags sse3 = feature_flags{ 0x0001'0000 }; inline constexpr feature_flags ssse3 = feature_flags{ 0x0002'0000 }; inline constexpr feature_flags sse4_1 = feature_flags{ 0x0004'0000 }; inline constexpr feature_flags sse4_2 = feature_flags{ 0x0008'0000 }; inline constexpr feature_flags xsave = feature_flags{ 0x0010'0000 }; inline constexpr feature_flags avx = feature_flags{ 0x0020'0000 }; inline constexpr feature_flags avx2 = feature_flags{ 0x0040'0000 }; inline constexpr feature_flags cx16 = feature_flags{ 0x0080'0000 }; inline constexpr feature_flags lahf = feature_flags{ 0x0100'0000 }; inline constexpr feature_flags popcnt = feature_flags{ 0x0200'0000 }; inline constexpr feature_flags bmi1 = feature_flags{ 0x0400'0000 }; inline constexpr feature_flags bmi2 = feature_flags{ 0x0800'0000 }; inline constexpr feature_flags f16c = feature_flags{ 0x1000'0000 }; inline constexpr feature_flags fma = feature_flags{ 0x2000'0000 }; inline constexpr feature_flags lzcnt = feature_flags{ 0x4000'0000 }; inline constexpr feature_flags movbe = feature_flags{ 0x8000'0000 }; } // namespace feature namespace mode { inline constexpr mode_flags base = mode_flags{ 0x00 }; inline constexpr mode_flags xmm128sse = mode_flags{ 0x01 }; inline constexpr mode_flags ymm256avx = mode_flags{ 0x02 }; } // namespace mode namespace featureset { inline constexpr feature_flags intel386 = feature::intel386; inline constexpr feature_flags intel486SX = featureset::intel386 | feature::intel486; inline constexpr feature_flags intel486DX = featureset::intel486SX | feature::fpu | feature::fsin; inline constexpr feature_flags intel586 = featureset::intel486DX | feature::cpuid | feature::tsc | feature::cx8; inline constexpr feature_flags intel586_mmx = featureset::intel586 | feature::mmx; inline constexpr feature_flags intel686 = featureset::intel586 | feature::cmov; inline constexpr feature_flags intel686_mmx = featureset::intel686 | feature::mmx; inline constexpr feature_flags intel686_sse = featureset::intel686_mmx | feature::mmxext | feature::fxsr | feature::sse; inline constexpr feature_flags intel686_sse2 = featureset::intel686_sse | feature::sse2; inline constexpr feature_flags intel786 = featureset::intel686_sse2; inline constexpr feature_flags amd64 = featureset::intel686_sse2; inline constexpr feature_flags amd64_v2 = featureset::amd64 | feature::cx16 | feature::lahf | feature::popcnt | feature::sse3 | feature::ssse3 | feature::sse4_1 | feature::sse4_2; inline constexpr feature_flags amd64_v3 = featureset::amd64_v2 | feature::xsave | feature::avx | feature::avx2 | feature::bmi1 | feature::bmi2 | feature::f16c | feature::fma | feature::lzcnt | feature::movbe; inline constexpr feature_flags msvc_x86_1998 = featureset::intel386 | feature::fpu | feature::fsin; inline constexpr feature_flags msvc_x86_2005 = featureset::intel486DX; inline constexpr feature_flags msvc_x86_2008 = featureset::intel586; inline constexpr feature_flags msvc_x86_sse = featureset::intel686_sse; inline constexpr feature_flags msvc_x86_sse2 = featureset::intel686_sse2; inline constexpr feature_flags msvc_x86_avx = featureset::intel686_sse2 | feature::x3dnowprefetch | feature::xsave | feature::avx; inline constexpr feature_flags msvc_x86_avx2 = featureset::intel686_sse2 | feature::x3dnowprefetch | feature::xsave | feature::avx | feature::avx2 | feature::fma | feature::bmi1; inline constexpr feature_flags msvc_amd64 = featureset::amd64; inline constexpr feature_flags msvc_amd64_avx = featureset::amd64 | feature::xsave | feature::avx; inline constexpr feature_flags msvc_amd64_avx2 = featureset::amd64 | feature::xsave | feature::avx | feature::avx2 | feature::fma | feature::bmi1; } // namespace featureset namespace modeset { inline constexpr mode_flags intel386 = mode::base; inline constexpr mode_flags intel486SX = mode::base; inline constexpr mode_flags intel486DX = mode::base; inline constexpr mode_flags intel586 = mode::base; inline constexpr mode_flags intel586_mmx = mode::base; inline constexpr mode_flags intel686 = mode::base; inline constexpr mode_flags intel686_mmx = mode::base; inline constexpr mode_flags intel686_sse = mode::base | mode::xmm128sse; inline constexpr mode_flags intel686_sse2 = mode::base | mode::xmm128sse; inline constexpr mode_flags intel786 = mode::base | mode::xmm128sse; inline constexpr mode_flags amd64 = mode::base | mode::xmm128sse; inline constexpr mode_flags amd64_v2 = mode::base | mode::xmm128sse; inline constexpr mode_flags amd64_v3 = mode::base | mode::xmm128sse | mode::ymm256avx; inline constexpr mode_flags msvc_x86_1998 = mode::base; inline constexpr mode_flags msvc_x86_2005 = mode::base; inline constexpr mode_flags msvc_x86_2008 = mode::base; inline constexpr mode_flags msvc_x86_sse = mode::base | mode::xmm128sse; inline constexpr mode_flags msvc_x86_sse2 = mode::base | mode::xmm128sse; inline constexpr mode_flags msvc_x86_avx = mode::base | mode::xmm128sse | mode::ymm256avx; inline constexpr mode_flags msvc_x86_avx2 = mode::base | mode::xmm128sse | mode::ymm256avx; inline constexpr mode_flags msvc_amd64 = mode::base | mode::xmm128sse; inline constexpr mode_flags msvc_amd64_avx = mode::base | mode::xmm128sse | mode::ymm256avx; inline constexpr mode_flags msvc_amd64_avx2 = mode::base | mode::xmm128sse | mode::ymm256avx; } // namespace modeset // clang-format on enum class vendor : uint8 { unknown = 0, AMD, Centaur, Cyrix, Intel, Transmeta, NSC, NexGen, Rise, SiS, UMC, VIA, DMnP, Zhaoxin, Hygon, Elbrus, MiSTer, bhyve, KVM, QEMU, HyperV, Parallels, VMWare, Xen, ACRN, QNX, }; // enum class vendor // clang-format off [[nodiscard]] MPT_CONSTEVAL feature_flags assumed_features() noexcept { feature_flags flags{}; #if MPT_ARCH_X86 || MPT_ARCH_AMD64 #ifdef MPT_ARCH_X86_I386 flags |= feature::intel386; #endif #ifdef MPT_ARCH_X86_FPU flags |= feature::fpu; #endif #ifdef MPT_ARCH_X86_FSIN flags |= feature::fsin; #endif #ifdef MPT_ARCH_X86_I486 flags |= feature::intel486; #endif #ifdef MPT_ARCH_X86_CPUID flags |= feature::cpuid; #endif #ifdef MPT_ARCH_X86_TSC flags |= feature::tsc; #endif #ifdef MPT_ARCH_X86_CX8 flags |= feature::cx8; #endif #ifdef MPT_ARCH_X86_CMOV flags |= feature::cmov; #endif #ifdef MPT_ARCH_X86_MMX flags |= feature::mmx; #endif #ifdef MPT_ARCH_X86_MMXEXT flags |= feature::mmxext; #endif #ifdef MPT_ARCH_X86_3DNOW flags |= feature::x3dnow; #endif #ifdef MPT_ARCH_X86_3DNOWEXT flags |= feature::x3dnowext; #endif #ifdef MPT_ARCH_X86_3DNOWPREFETCH flags |= feature::x3dnowprefetch; #endif #ifdef MPT_ARCH_X86_FXSR flags |= feature::fxsr; #endif #ifdef MPT_ARCH_X86_SSE flags |= feature::sse; #endif #ifdef MPT_ARCH_X86_SSE2 flags |= feature::sse2; #endif #ifdef MPT_ARCH_X86_SSE3 flags |= feature::sse3; #endif #ifdef MPT_ARCH_X86_SSSE3 flags |= feature::ssse3; #endif #ifdef MPT_ARCH_X86_SSE4_1 flags |= feature::sse4_1; #endif #ifdef MPT_ARCH_X86_SSE4_2 flags |= feature::sse4_2; #endif #ifdef MPT_ARCH_X86_XSAVE flags |= feature::xsave; #endif #ifdef MPT_ARCH_X86_AVX flags |= feature::avx; #endif #ifdef MPT_ARCH_X86_AVX2 flags |= feature::avx2; #endif #ifdef MPT_ARCH_X86_CX16 flags |= feature::cx16; #endif #ifdef MPT_ARCH_X86_LAHF flags |= feature::lahf; #endif #ifdef MPT_ARCH_X86_POPCNT flags |= feature::popcnt; #endif #ifdef MPT_ARCH_X86_BMI1 flags |= feature::bmi1; #endif #ifdef MPT_ARCH_X86_BMI2 flags |= feature::bmi2; #endif #ifdef MPT_ARCH_X86_F16C flags |= feature::f16c; #endif #ifdef MPT_ARCH_X86_FMA flags |= feature::fma; #endif #ifdef MPT_ARCH_X86_LZCNT flags |= feature::lzcnt; #endif #ifdef MPT_ARCH_X86_MOVBE flags |= feature::movbe; #endif #endif // MPT_ARCH_X86 || MPT_ARCH_AMD64 return flags; } // clang-format on // clang-format off [[nodiscard]] MPT_CONSTEVAL mode_flags assumed_modes() noexcept { mode_flags flags{}; #if MPT_ARCH_X86 || MPT_ARCH_AMD64 #ifdef MPT_ARCH_X86_SSE flags |= mode::xmm128sse; #endif #ifdef MPT_ARCH_X86_AVX flags |= mode::ymm256avx; #endif #endif // MPT_ARCH_X86 || MPT_ARCH_AMD64 return flags; } // clang-format on template struct fixed_string { std::array m_data = {}; [[nodiscard]] constexpr const char & operator[](std::size_t i) const noexcept { return m_data[i]; } [[nodiscard]] constexpr char & operator[](std::size_t i) noexcept { return m_data[i]; } [[nodiscard]] constexpr std::size_t size() const noexcept { return m_data.size(); } [[nodiscard]] constexpr const char * data() const noexcept { return m_data.data(); } [[nodiscard]] constexpr char * data() noexcept { return m_data.data(); } [[nodiscard]] constexpr const char * begin() const noexcept { return m_data.data(); } [[nodiscard]] constexpr char * begin() noexcept { return m_data.data(); } [[nodiscard]] constexpr const char * end() const noexcept { return m_data.data() + m_data.size(); } [[nodiscard]] constexpr char * end() noexcept { return m_data.data() + m_data.size(); } [[nodiscard]] constexpr operator std::string_view() const noexcept { #if MPT_CXX_AT_LEAST(20) && !defined(MPT_COMPILER_QUIRK_NO_STRING_VIEW_ITERATOR_CTOR) return std::string_view(m_data.begin(), std::find(m_data.begin(), m_data.end(), '\0')); #else return std::string_view(m_data.data(), std::find(m_data.begin(), m_data.end(), '\0') - m_data.begin()); #endif } template [[nodiscard]] friend MPT_CONSTEXPR20_FUN auto operator+(fixed_string a, fixed_string b) -> fixed_string { fixed_string result; std::copy(a.begin(), a.end(), result.data() + 0); std::copy(b.begin(), b.end(), result.data() + N); return result; } [[nodiscard]] inline explicit operator std::string() const { return std::string(m_data.begin(), std::find(m_data.begin(), m_data.end(), '\0')); } }; struct cpu_info { private: feature_flags Features{}; mode_flags Modes{}; uint32 CPUID = 0; vendor Vendor = vendor::unknown; uint16 Family = 0; uint8 Model = 0; uint8 Stepping = 0; fixed_string<12> VendorID; fixed_string<48> BrandID; bool Virtualized = false; fixed_string<12> HypervisorVendor; fixed_string<4> HypervisorInterface; #if !MPT_ARCH_AMD64 bool LongMode = false; #endif // !MPT_ARCH_AMD64 public: [[nodiscard]] MPT_CONSTEXPRINLINE bool operator[](feature_flags query_features) const noexcept { return ((Features & query_features) == query_features); } [[nodiscard]] MPT_CONSTEXPRINLINE bool has_features(feature_flags query_features) const noexcept { return ((Features & query_features) == query_features); } [[nodiscard]] MPT_CONSTEXPRINLINE feature_flags get_features() const noexcept { return Features; } [[nodiscard]] MPT_CONSTEXPRINLINE bool operator[](mode_flags query_modes) const noexcept { return ((Modes & query_modes) == query_modes); } [[nodiscard]] MPT_CONSTEXPRINLINE bool enabled_modes(mode_flags query_modes) const noexcept { return ((Modes & query_modes) == query_modes); } [[nodiscard]] MPT_CONSTEXPRINLINE mode_flags get_modes() const noexcept { return Modes; } [[nodiscard]] MPT_CONSTEXPRINLINE uint32 get_cpuid() const noexcept { return CPUID; } [[nodiscard]] MPT_CONSTEXPRINLINE vendor get_vendor() const noexcept { return Vendor; } [[nodiscard]] MPT_CONSTEXPRINLINE uint16 get_family() const noexcept { return Family; } [[nodiscard]] MPT_CONSTEXPRINLINE uint8 get_model() const noexcept { return Model; } [[nodiscard]] MPT_CONSTEXPRINLINE uint8 get_stepping() const noexcept { return Stepping; } [[nodiscard]] inline std::string get_vendor_string() const { return std::string(VendorID); } [[nodiscard]] inline std::string get_brand_string() const { return std::string(BrandID); } [[nodiscard]] MPT_CONSTEXPRINLINE bool is_virtual() const noexcept { return Virtualized; } [[nodiscard]] MPT_CONSTEXPRINLINE bool can_long_mode() const noexcept { #if !MPT_ARCH_AMD64 return LongMode; #else // MPT_ARCH_AMD64 return true; #endif // !MPT_ARCH_AMD64 } private: #if MPT_ARCH_X86 || MPT_ARCH_AMD64 struct cpuid_result { uint32 a = 0; uint32 b = 0; uint32 c = 0; uint32 d = 0; [[nodiscard]] MPT_CONSTEXPR20_FUN fixed_string<4> as_text4() const noexcept { fixed_string<4> result; result[0 + 0] = static_cast((a >> 0) & 0xff); result[0 + 1] = static_cast((a >> 8) & 0xff); result[0 + 2] = static_cast((a >> 16) & 0xff); result[0 + 3] = static_cast((a >> 24) & 0xff); return result; } [[nodiscard]] MPT_CONSTEXPR20_FUN fixed_string<12> as_text12bcd() const noexcept { fixed_string<12> result; result[0 + 0] = static_cast((b >> 0) & 0xff); result[0 + 1] = static_cast((b >> 8) & 0xff); result[0 + 2] = static_cast((b >> 16) & 0xff); result[0 + 3] = static_cast((b >> 24) & 0xff); result[4 + 0] = static_cast((c >> 0) & 0xff); result[4 + 1] = static_cast((c >> 8) & 0xff); result[4 + 2] = static_cast((c >> 16) & 0xff); result[4 + 3] = static_cast((c >> 24) & 0xff); result[8 + 0] = static_cast((d >> 0) & 0xff); result[8 + 1] = static_cast((d >> 8) & 0xff); result[8 + 2] = static_cast((d >> 16) & 0xff); result[8 + 3] = static_cast((d >> 24) & 0xff); return result; } [[nodiscard]] MPT_CONSTEXPR20_FUN fixed_string<12> as_text12bdc() const noexcept { fixed_string<12> result; result[0 + 0] = static_cast((b >> 0) & 0xff); result[0 + 1] = static_cast((b >> 8) & 0xff); result[0 + 2] = static_cast((b >> 16) & 0xff); result[0 + 3] = static_cast((b >> 24) & 0xff); result[4 + 0] = static_cast((d >> 0) & 0xff); result[4 + 1] = static_cast((d >> 8) & 0xff); result[4 + 2] = static_cast((d >> 16) & 0xff); result[4 + 3] = static_cast((d >> 24) & 0xff); result[8 + 0] = static_cast((c >> 0) & 0xff); result[8 + 1] = static_cast((c >> 8) & 0xff); result[8 + 2] = static_cast((c >> 16) & 0xff); result[8 + 3] = static_cast((c >> 24) & 0xff); return result; } [[nodiscard]] MPT_CONSTEXPR20_FUN fixed_string<16> as_text16() const noexcept { fixed_string<16> result; result[0 + 0] = static_cast((a >> 0) & 0xff); result[0 + 1] = static_cast((a >> 8) & 0xff); result[0 + 2] = static_cast((a >> 16) & 0xff); result[0 + 3] = static_cast((a >> 24) & 0xff); result[4 + 0] = static_cast((b >> 0) & 0xff); result[4 + 1] = static_cast((b >> 8) & 0xff); result[4 + 2] = static_cast((b >> 16) & 0xff); result[4 + 3] = static_cast((b >> 24) & 0xff); result[8 + 0] = static_cast((c >> 0) & 0xff); result[8 + 1] = static_cast((c >> 8) & 0xff); result[8 + 2] = static_cast((c >> 16) & 0xff); result[8 + 3] = static_cast((c >> 24) & 0xff); result[12 + 0] = static_cast((d >> 0) & 0xff); result[12 + 1] = static_cast((d >> 8) & 0xff); result[12 + 2] = static_cast((d >> 16) & 0xff); result[12 + 3] = static_cast((d >> 24) & 0xff); return result; } }; #if MPT_COMPILER_MSVC || MPT_COMPILER_GCC || MPT_COMPILER_CLANG [[nodiscard]] static cpuid_result cpuid(uint32 function) noexcept { #if MPT_COMPILER_MSVC cpuid_result result; int CPUInfo[4]{}; __cpuid(CPUInfo, function); result.a = CPUInfo[0]; result.b = CPUInfo[1]; result.c = CPUInfo[2]; result.d = CPUInfo[3]; return result; #elif MPT_COMPILER_GCC || (MPT_COMPILER_CLANG && !MPT_ARCH_AMD64) || MPT_CLANG_AT_LEAST(13, 0, 0) // cpuid_result result; unsigned int regeax{}; unsigned int regebx{}; unsigned int regecx{}; unsigned int regedx{}; __cpuid(function, regeax, regebx, regecx, regedx); result.a = regeax; result.b = regebx; result.c = regecx; result.d = regedx; return result; #elif MPT_COMPILER_CLANG && MPT_ARCH_AMD64 // cpuid_result result; unsigned int a{}; unsigned int b{}; unsigned int c{}; unsigned int d{}; // clang-format off __asm__ __volatile__ ( "xchgq %%rbx,%q1 \n\t" "cpuid \n\t" "xchgq %%rbx,%q1 \n\t" : "=a" (a), "=r" (b), "=c" (c), "=d" (d) : "0" (function)); // clang-format on result.a = a; result.b = b; result.c = c; result.d = d; return result; #elif 0 cpuid_result result; unsigned int a{}; unsigned int b{}; unsigned int c{}; unsigned int d{}; // clang-format off __asm__ __volatile__ ( "cpuid \n\t" : "=a" (a), "=b" (b), "=c" (c), "=d" (d) : "0" (function)); // clang-format on result.a = a; result.b = b; result.c = c; result.d = d; return result; #else // MPT_COMPILER return cpuid_result result{}; #endif // MPT_COMPILER } [[nodiscard]] static cpuid_result cpuidex(uint32 function_a, uint32 function_c) noexcept { #if MPT_COMPILER_MSVC cpuid_result result; int CPUInfo[4]{}; __cpuidex(CPUInfo, function_a, function_c); result.a = CPUInfo[0]; result.b = CPUInfo[1]; result.c = CPUInfo[2]; result.d = CPUInfo[3]; return result; #elif MPT_COMPILER_GCC || (MPT_COMPILER_CLANG && !MPT_ARCH_AMD64) || MPT_CLANG_AT_LEAST(13, 0, 0) // cpuid_result result; unsigned int regeax{}; unsigned int regebx{}; unsigned int regecx{}; unsigned int regedx{}; __cpuid_count(function_a, function_c, regeax, regebx, regecx, regedx); result.a = regeax; result.b = regebx; result.c = regecx; result.d = regedx; return result; #elif MPT_COMPILER_CLANG && MPT_ARCH_AMD64 // cpuid_result result; unsigned int a{}; unsigned int b{}; unsigned int c{}; unsigned int d{}; // clang-format off __asm__ __volatile__ ( "xchgq %%rbx,%q1 \n\t" "cpuid \n\t" "xchgq %%rbx,%q1 \n\t" : "=a" (a), "=r" (b), "=c" (c), "=d" (d) : "0" (function_a), "2" (function_c)); // clang-format on result.a = a; result.b = b; result.c = c; result.d = d; return result; #elif 0 cpuid_result result; unsigned int a{}; unsigned int b{}; unsigned int c{}; unsigned int d{}; // clang-format off __asm__ __volatile__ ( "cpuid \n\t" : "=a" (a), "=b" (b), "=c" (c), "=d" (d) : "0" (function_a), "2" (function_c)); // clang-format on result.a = a; result.b = b; result.c = c; result.d = d; return result; #else // MPT_COMPILER return cpuid_result result{}; #endif // MPT_COMPILER } #if MPT_MODE_KERNEL [[nodiscard]] static uint32 read_cr4() noexcept { return __readcr4(); } #endif // MPT_MODE_KERNEL [[nodiscard]] static uint64 read_xcr(uint32 num) noexcept { #if MPT_COMPILER_MSVC return _xgetbv(num); #elif MPT_COMPILER_GCC || MPT_COMPILER_CLANG uint32 param_ecx = num; uint32 result_eax = 0; uint32 result_edx = 0; // clang-format off __asm__ __volatile__("xgetbv" : "=a" (result_eax), "=d" (result_edx) : "c" (param_ecx)); // clang-format on return static_cast(result_eax) + (static_cast(result_edx) << 32); #else return _xgetbv(num); #endif } #endif // MPT_COMPILER_MSVC || MPT_COMPILER_GCC || MPT_COMPILER_CLANG private: #if MPT_ARCH_X86 || !MPT_ARCH_X86_AMD64_FAST_DETECT #if MPT_OS_DJGPP [[nodiscard]] static uint8 detect_cpu_level() noexcept { #if 0 uint8 result = 0; __dpmi_version_ret dpmi_version{}; if (__dpmi_get_version(&dpmi_version) == 0) { result = dpmi_version.cpu; } return result; #else uint8 result = 0; __dpmi_regs regs{}; regs.x.ax = 0x0400; if (__dpmi_int(0x31, ®s) == 0) { unsigned int cf = (regs.x.flags >> 0) & 1u; if (cf == 0) { result = regs.h.cl; } } return result; #endif } #endif // MPT_OS_DJGPP [[nodiscard]] static uint8 detect_fpu_level() noexcept { #if MPT_OS_DJGPP #if 1 uint8 result = 0; int coprocessor_status = __dpmi_get_coprocessor_status(); if (coprocessor_status == -1) { // error = __dpmi_error return 0; } result = (static_cast(coprocessor_status) & 0x00f0u) >> 4; return result; #else uint8 result = 0; __dpmi_regs regs{}; regs.x.ax = 0x0e00; if (__dpmi_int(0x31, ®s) == 0) { unsigned int cf = (regs.x.flags >> 0) & 1u; if (cf == 0) { result = (regs.x.ax & 0x00f0u) >> 4; } } return result; #endif #elif MPT_OS_WINDOWS uint8 result = 0; #if MPT_WINNT_AT_LEAST(MPT_WIN_NT4) if (mpt::osinfo::windows::Version::Current().IsAtLeast(mpt::osinfo::windows::Version::Win2000)) { if (IsProcessorFeaturePresent(PF_FLOATING_POINT_EMULATED) == 0) { result = 3; } } else { // logic is inverted on NT4 if (IsProcessorFeaturePresent(PF_FLOATING_POINT_EMULATED) != 0) { result = 3; } } #else if ((assumed_features() & feature::fpu) && (assumed_features() & feature::fsin)) { result = 3; } else if (assumed_features() & feature::fpu) { result = 2; } else { result = 0; } #endif return result; #elif MPT_COMPILER_MSVC && MPT_MODE_KERNEL uint8 result = 0; const std::size_t cr0 = __readcr0(); if (!(cr0 & (1u << 2))) { // EM result = 2; if (cr0 & (1u << 4)) { // ET result = 3; } } return result; #else uint8 result = 0; if ((assumed_features() & feature::fpu) && (assumed_features() & feature::fsin)) { result = 3; } else if (assumed_features() & feature::fpu) { result = 2; } else { result = 0; } return result; #endif } #if MPT_COMPILER_MSVC || MPT_COMPILER_GCC || MPT_COMPILER_CLANG [[nodiscard]] static bool can_toggle_eflags(std::size_t mask) noexcept { std::size_t eflags_old = __readeflags(); std::size_t eflags_flipped = eflags_old ^ mask; __writeeflags(eflags_flipped); std::size_t eflags_testchanged = __readeflags(); __writeeflags(eflags_old); return ((eflags_testchanged ^ eflags_old) & mask) == mask; } #endif [[nodiscard]] static bool can_toggle_eflags_ac() noexcept { #if MPT_COMPILER_MSVC || MPT_COMPILER_GCC || MPT_COMPILER_CLANG return can_toggle_eflags(0x0004'0000); #else // MPT_COMPILER return (assumed_features() & feature::intel486) != 0; #endif // MPT_COMPILER } [[nodiscard]] static bool can_toggle_eflags_id() noexcept { #if MPT_COMPILER_MSVC || MPT_COMPILER_GCC || MPT_COMPILER_CLANG return can_toggle_eflags(0x0020'0000); #else // MPT_COMPILER return (assumed_features() & feature::tsc) != 0; #endif // MPT_COMPILER } [[nodiscard]] MPT_NOINLINE static bool detect_nexgen() noexcept { #if MPT_ARCH_X86 && MPT_COMPILER_MSVC uint8 result = 0; // clang-format off _asm { mov eax, 0x5555 xor edx, edx mov ecx, 2 clc div ecx jz found jmp done found: mov result, 1 jmp done done: } // clang-format on return (result != 0); #elif MPT_ARCH_X86 && (MPT_COMPILER_GCC || MPT_COMPILER_CLANG) unsigned int result = 0; // clang-format off __asm__ __volatile( "movl $0x5555, %%eax\n" "xorl %%edx, %%edx\n" "movl $2, %%ecx\n" "clc\n" "divl %%ecx\n" "jz detect_nexgen_found\n" "movl $0, %%eax\n" "jmp detect_nexgen_done\n" "detect_nexgen_found:\n" "movl $0, %%eax\n" "jmp detect_nexgen_done\n" "detect_nexgen_done:\n" : "=a" (result) : : "%ebx", "%ecx" ); // clang-format on return (result != 0); #else // assume false return false; #endif } [[nodiscard]] MPT_NOINLINE static bool detect_cyrix() noexcept { #if MPT_ARCH_X86 && MPT_COMPILER_MSVC uint8 result = 0; // clang-format off _asm { xor ax, ax sahf mov ax, 5 mov bx, 2 div bl lahf cmp ah, 2 jne not_cyrix mov result, 1 not_cyrix: } // clang-format on return (result != 0); #elif MPT_ARCH_X86 && (MPT_COMPILER_GCC || MPT_COMPILER_CLANG) unsigned int result = 0; // clang-format off __asm__ __volatile( "xor %%ax, %%ax\n" "sahf\n" "movw $5, %%ax\n" "movw $2, %%bx\n" "divb %%bl\n" "lahf\n" "cmpb $2, %%ah\n" "jne detect_cyrix_done_not_cyrix\n" "movl $1, %%eax\n" "jmp detect_cyrix_done\n" "detect_cyrix_done_not_cyrix:\n" "movl $0, %%eax\n" "jmp detect_cyrix_done\n" "detect_cyrix_done:\n" : "=a" (result) : : "%ebx" ); // clang-format on return (result != 0); #else // assume false return false; #endif } [[nodiscard]] static uint16 read_cyrix_id() noexcept { #if MPT_OS_DJGPP uint16 result = 0; outportb(0x22, 0x00); result |= static_cast(inportb(0x23)) << 8; outportb(0x22, 0x01); result |= static_cast(inportb(0x23)) << 0; outportb(0x22, 0x00); inportb(0x23); return result; #elif MPT_COMPILER_MSVC && MPT_MODE_KERNEL uint16 result = 0; __outbyte(0x22, 0x00); result |= static_cast(__inbyte(0x23)) << 8; __outbyte(0x22, 0x01); result |= static_cast(__inbyte(0x23)) << 0; __outbyte(0x22, 0x00); __inbyte(0x23); return result; #else return 0x00'00; #endif } #endif // MPT_ARCH_X86 #endif // MPT_ARCH_X86 || MPT_ARCH_AMD64 public: cpu_info() noexcept { #if MPT_ARCH_X86 || MPT_ARCH_AMD64 #if MPT_COMPILER_MSVC || MPT_COMPILER_GCC || MPT_COMPILER_CLANG #if MPT_ARCH_X86 || !MPT_ARCH_X86_AMD64_FAST_DETECT Features |= featureset::intel386; if (can_toggle_eflags_ac()) { Features |= feature::intel486; } if (can_toggle_eflags_id()) { Features |= feature::cpuid; } if (!Features.supports(feature::cpuid)) { // without cpuid if (!Features.supports(feature::intel486)) { // 386 const uint8 fpu_level = detect_fpu_level(); if (fpu_level >= 2) { Features |= feature::fpu; } if (fpu_level >= 3) { Features |= feature::fsin; } if (detect_nexgen()) { Vendor = vendor::NexGen; } else { Vendor = vendor::unknown; } } else { // 486+ const uint8 fpu_level = detect_fpu_level(); if (fpu_level >= 2) { Features |= feature::fpu; } if (fpu_level >= 3) { Features |= feature::fsin; } if (detect_cyrix()) { Vendor = vendor::Cyrix; uint16 id = read_cyrix_id(); if (id <= 0x07'00) { // Cx486SLC / Cx486DLC Family = 3; Model = static_cast((id & 0xff'00) >> 8); Stepping = static_cast(id & 0x00'ff); } else if ((0x10'00 <= id) && (id <= 0x13'00)) { // Cx486S Family = 4; Model = static_cast((id & 0xff'00) >> 8); Stepping = static_cast(id & 0x00'ff); } else if ((0x1a'00 <= id) && (id <= 0x1f'00)) { // Cx486DX Family = 4; Model = static_cast((id & 0xff'00) >> 8); Stepping = static_cast(id & 0x00'ff); } else if ((0x28'00 <= id) && (id <= 0x2e'00)) { // Cx5x86 Family = 4; Model = static_cast((id & 0xff'00) >> 8); Stepping = static_cast(id & 0x00'ff); } else if ((0x40'00 <= id) && (id <= 0x4f'00)) { // MediaGx / MediaGXm Family = 4; Model = 4; Stepping = static_cast(id & 0x00'ff); } else if ((0x30'00 <= id) && (id <= 0x34'00)) { // Cx6x86 / Cx6x86L Family = 5; Model = 2; Stepping = static_cast(id & 0x00'ff); if ((id & 0x00'ff) > 0x21) { // Cx6x86L Features |= feature::cx8; } } else if ((0x50'00 <= id) && (id <= 0x5f'00)) { // Cx6x86MX Family = 6; Model = 0; Stepping = static_cast(id & 0x00'ff); Features |= feature::cx8; Features |= feature::tsc; Features |= feature::mmx; Features |= feature::cmov; } } else { Vendor = vendor::unknown; } } } #elif MPT_ARCH_AMD64 Features |= featureset::amd64; Modes |= modeset::amd64; #endif // MPT_ARCH bool have_osxsave = false; if (Features.supports(feature::cpuid)) { // with cpuid cpuid_result VendorString = cpuid(0x0000'0000u); VendorID = VendorString.as_text12bdc(); if (VendorID == std::string_view(" ")) { Vendor = vendor::unknown; } else if (VendorID == std::string_view("AMDisbetter!")) { Vendor = vendor::AMD; } else if (VendorID == std::string_view("AuthenticAMD")) { Vendor = vendor::AMD; } else if (VendorID == std::string_view("CentaurHauls")) { Vendor = vendor::Centaur; } else if (VendorID == std::string_view("CyrixInstead")) { Vendor = vendor::Cyrix; } else if (VendorID == std::string_view("GenuineIntel")) { Vendor = vendor::Intel; } else if (VendorID == std::string_view("TransmetaCPU")) { Vendor = vendor::Transmeta; } else if (VendorID == std::string_view("GenuineTMx86")) { Vendor = vendor::Transmeta; } else if (VendorID == std::string_view("Geode by NSC")) { Vendor = vendor::NSC; } else if (VendorID == std::string_view("NexGenDriven")) { Vendor = vendor::NexGen; } else if (VendorID == std::string_view("RiseRiseRise")) { Vendor = vendor::Rise; } else if (VendorID == std::string_view("SiS SiS SiS ")) { Vendor = vendor::SiS; } else if (VendorID == std::string_view("UMC UMC UMC ")) { Vendor = vendor::UMC; } else if (VendorID == std::string_view("VIA VIA VIA ")) { Vendor = vendor::VIA; } else if (VendorID == std::string_view("Vortex86 SoC")) { Vendor = vendor::DMnP; } else if (VendorID == std::string_view(" Shanghai ")) { Vendor = vendor::Zhaoxin; } else if (VendorID == std::string_view("HygonGenuine")) { Vendor = vendor::Hygon; } else if (VendorID == std::string_view("E2K MACHINE")) { Vendor = vendor::Elbrus; } else if (VendorID == std::string_view("MiSTer AO486")) { Vendor = vendor::MiSTer; } else if (VendorID == std::string_view("bhyve bhyve ")) { Vendor = vendor::bhyve; } else if (VendorID == std::string_view(" KVMKVMKVM ")) { Vendor = vendor::KVM; } else if (VendorID == std::string_view("TCGTCGTCGTCG")) { Vendor = vendor::QEMU; } else if (VendorID == std::string_view("Microsoft Hv")) { Vendor = vendor::HyperV; } else if (VendorID == std::string_view(" lrpepyh vr")) { Vendor = vendor::Parallels; } else if (VendorID == std::string_view("VMwareVMware")) { Vendor = vendor::VMWare; } else if (VendorID == std::string_view("XenVMMXenVMM")) { Vendor = vendor::Xen; } else if (VendorID == std::string_view("ACRNACRNACRN")) { Vendor = vendor::ACRN; } else if (VendorID == std::string_view(" QNXQVMBSQG ")) { Vendor = vendor::QNX; } // Cyrix 6x86 and 6x86MX do not specify the value returned in eax. // They both support 0x00000001u however. if ((VendorString.a >= 0x0000'0001u) || (Vendor == vendor::Cyrix)) { cpuid_result StandardFeatureFlags = cpuid(0x0000'0001u); CPUID = StandardFeatureFlags.a; // clang-format off uint32 BaseStepping = (StandardFeatureFlags.a >> 0) & 0x0f; uint32 BaseModel = (StandardFeatureFlags.a >> 4) & 0x0f; uint32 BaseFamily = (StandardFeatureFlags.a >> 8) & 0x0f; uint32 ExtModel = (StandardFeatureFlags.a >> 16) & 0x0f; uint32 ExtFamily = (StandardFeatureFlags.a >> 20) & 0xff; // clang-format on if (BaseFamily == 0xf) { Family = static_cast(ExtFamily + BaseFamily); } else { Family = static_cast(BaseFamily); } if ((Vendor == vendor::AMD) && (BaseFamily == 0xf)) { Model = static_cast((ExtModel << 4) | (BaseModel << 0)); } else if ((Vendor == vendor::Centaur) && (BaseFamily >= 0x6)) { // Newer Zhaoxin CPUs use extended family also with BaseFamily=7. Model = static_cast((ExtModel << 4) | (BaseModel << 0)); } else if ((BaseFamily == 0x6) || (BaseFamily == 0xf)) { Model = static_cast((ExtModel << 4) | (BaseModel << 0)); } else { Model = static_cast(BaseModel); } Stepping = static_cast(BaseStepping); // clang-format off Features |= (StandardFeatureFlags.d & (1u << 0)) ? (feature::fpu | feature::fsin) : feature::none; Features |= (StandardFeatureFlags.d & (1u << 4)) ? (feature::tsc) : feature::none; Features |= (StandardFeatureFlags.d & (1u << 8)) ? (feature::cx8) : feature::none; Features |= (StandardFeatureFlags.d & (1u << 15)) ? (feature::cmov) : feature::none; Features |= (StandardFeatureFlags.d & (1u << 23)) ? (feature::mmx) : feature::none; Features |= (StandardFeatureFlags.d & (1u << 24)) ? (feature::fxsr) : feature::none; Features |= (StandardFeatureFlags.d & (1u << 25)) ? (feature::sse | feature::mmxext) : feature::none; Features |= (StandardFeatureFlags.d & (1u << 26)) ? (feature::sse2) : feature::none; Features |= (StandardFeatureFlags.c & (1u << 0)) ? (feature::sse3) : feature::none; Features |= (StandardFeatureFlags.c & (1u << 9)) ? (feature::ssse3) : feature::none; Features |= (StandardFeatureFlags.c & (1u << 12)) ? (feature::fma) : feature::none; Features |= (StandardFeatureFlags.c & (1u << 13)) ? (feature::cx16) : feature::none; Features |= (StandardFeatureFlags.c & (1u << 19)) ? (feature::sse4_1) : feature::none; Features |= (StandardFeatureFlags.c & (1u << 20)) ? (feature::sse4_2) : feature::none; Features |= (StandardFeatureFlags.c & (1u << 22)) ? (feature::movbe) : feature::none; Features |= (StandardFeatureFlags.c & (1u << 23)) ? (feature::popcnt) : feature::none; Features |= (StandardFeatureFlags.c & (1u << 26)) ? (feature::xsave) : feature::none; Features |= (StandardFeatureFlags.c & (1u << 28)) ? (feature::avx) : feature::none; Features |= (StandardFeatureFlags.c & (1u << 29)) ? (feature::f16c) : feature::none; // clang-format on if (StandardFeatureFlags.c & (1u << 27)) { have_osxsave = true; } if (StandardFeatureFlags.c & (1u << 31)) { Virtualized = true; } } if ((VendorString.a >= 0x0000'0007u) && Features.supports(feature::avx)) { // Some Skylake Celerons wrongly report BMI1 even though they do not support it. // They also do not support AVX. // Guard BMI1/BMI2 with AVX, as all known CPUs with BMI1 also support AVX. // See . cpuid_result ExtendedFeatures = cpuidex(0x0000'0007u, 0x0000'0000u); // clang-format off Features |= (ExtendedFeatures.b & (1u << 3)) ? (feature::bmi1) : feature::none; Features |= (ExtendedFeatures.b & (1u << 5)) ? (feature::avx2) : feature::none; Features |= (ExtendedFeatures.b & (1u << 8)) ? (feature::bmi2) : feature::none; // clang-format on } // 3DNow! manual recommends to just execute 0x8000'0000u. // It is totally unknown how earlier CPUs from other vendors // would behave. // Thus we only execute 0x80000000u on other vendors CPUs for the earliest // that we found it documented for and that actually supports 3DNow!. bool ecpuid = false; bool x3dnowknown = false; if (Vendor == vendor::Intel) { ecpuid = true; } else if (Vendor == vendor::AMD) { if ((Family > 5) || ((Family == 5) && (Model >= 8))) { // >= K6-2 (K6 = Family 5, K6-2 = Model 8) // Not sure if earlier AMD CPUs support 0x80000000u. // AMD 5k86 and AMD K5 manuals do not mention it. ecpuid = true; x3dnowknown = true; } } else if (Vendor == vendor::Centaur) { // Centaur (IDT WinChip or VIA C3) if (Family == 5) { // IDT if (Model >= 8) { // >= WinChip 2 ecpuid = true; x3dnowknown = true; } } else if (Family >= 6) { // VIA if ((Family >= 7) || ((Family == 6) && (Model >= 7))) { // >= C3 Samuel 2 ecpuid = true; x3dnowknown = true; } } } else if (Vendor == vendor::Cyrix) { // Cyrix // 6x86 : 5.2.x // 6x86L : 5.2.x // MediaGX : 4.4.x // 6x86MX : 6.0.x // MII : 6.0.x // MediaGXm: 5.4.x // well, doh ... if ((Family == 5) && (Model >= 4)) { // Cyrix MediaGXm ecpuid = true; x3dnowknown = true; } } else if (Vendor == vendor::NSC) { // National Semiconductor if ((Family > 5) || ((Family == 5) && (Model >= 5))) { // >= Geode GX2 ecpuid = true; x3dnowknown = true; } } else { // Intel specification allows 0x8000'0000u on earlier CPUs, // thus we execute it on unknown vendors. ecpuid = true; } if (ecpuid) { cpuid_result ExtendedVendorString = cpuid(0x8000'0000u); if ((ExtendedVendorString.a & 0xffff'0000u) == 0x8000'0000u) { if (ExtendedVendorString.a >= 0x8000'0001u) { // clang-format off cpuid_result ExtendedFeatureFlags = cpuid(0x8000'0001u); #if !MPT_ARCH_AMD64 if (ExtendedFeatureFlags.d & (1u << 29)) { LongMode = true; } #endif // !MPT_ARCH_AMD64 Features |= (ExtendedFeatureFlags.c & (1u << 0)) ? (feature::lahf) : feature::none; Features |= (ExtendedFeatureFlags.c & (1u << 5)) ? (feature::lzcnt) : feature::none; if (x3dnowknown) { Features |= (ExtendedFeatureFlags.d & (1u << 31)) ? (feature::x3dnow) : feature::none; } if (Vendor == vendor::AMD) { Features |= (ExtendedFeatureFlags.d & (1u << 22)) ? (feature::mmxext) : feature::none; Features |= (ExtendedFeatureFlags.d & (1u << 30)) ? (feature::x3dnowext) : feature::none; Features |= (ExtendedFeatureFlags.c & (1u << 5)) ? (feature::popcnt) : feature::none; Features |= (ExtendedFeatureFlags.c & (1u << 8)) ? (feature::x3dnowprefetch) : feature::none; } // clang-format on } if (ExtendedVendorString.a >= 0x8000'0004u) { BrandID = cpuid(0x8000'0002u).as_text16() + cpuid(0x8000'0003u).as_text16() + cpuid(0x8000'0004u).as_text16(); } } } #if MPT_ARCH_AMD64 // clang-format off Modes |= mode::xmm128sse; const bool have_xsave = #ifdef MPT_ARCH_X86_XSAVE true; #else Features.supports(feature::xsave); #endif MPT_MAYBE_CONSTANT_IF (have_xsave && have_osxsave) { const uint64 xcr0 = read_xcr(0x0000'0000u); Modes |= (xcr0 & (1ull << 2)) ? mode::ymm256avx : mode::base; } // clang-format on #else // !MPT_ARCH_AMD64 // clang-format off const bool have_xsave = #ifdef MPT_ARCH_X86_XSAVE true; #else Features.supports(feature::xsave); #endif MPT_MAYBE_CONSTANT_IF (have_xsave && have_osxsave) { const uint64 xcr0 = read_xcr(0x0000'0000u); Modes |= (xcr0 & (1ull << 1)) ? mode::xmm128sse : mode::base; Modes |= (xcr0 & (1ull << 2)) ? mode::ymm256avx : mode::base; } else { const bool have_fxsr = #ifdef MPT_ARCH_X86_FXSR true; #else Features.supports(feature::fxsr); #endif MPT_MAYBE_CONSTANT_IF (have_fxsr) { #if MPT_MODE_KERNEL const uint32 cr4 = read_cr4(); Modes |= (cr4 & (1u << 9)) ? mode::xmm128sse : mode::base; #else // !MPT_MODE_KERNEL // There is no way for user-mode code to check for SSE enabled in CR4. // Assume based on FXSR and SSE and platform heuristics. // Avoid assumption on DOS. #if MPT_OS_WINDOWS // get from platform API #if MPT_OS_WINDOWS_WINNT #ifdef PF_XMMI_INSTRUCTIONS_AVAILABLE Modes |= (IsProcessorFeaturePresent(PF_XMMI_INSTRUCTIONS_AVAILABLE) != 0) ? (mode::xmm128sse) : mode::base; #endif #elif MPT_WIN9X_AT_LEAST(MPT_WIN_WIN98) const bool have_sse = #ifdef MPT_ARCH_X86_SSE true; #else Features.supports(feature::sse); #endif MPT_MAYBE_CONSTANT_IF (have_sse) { Modes |= mode::xmm128sse; } #else // MPT_OS_WINDOWS // nothing #endif // MPT_OS_WINDOWS #elif MPT_OS_DJGPP || MPT_OS_UNKNOWN // avoid SSE on DOS and unknown systems, // however if we directly targeting >= SSE, then assume it's activated because the process will crash otherwise anyway #ifdef MPT_ARCH_X86_SSE Modes |= mode::xmm128sse; #endif #else // MPT_OS // assume based on FXSR and SSE const bool have_sse = #ifdef MPT_ARCH_X86_SSE true; #else Features.supports(feature::sse); #endif MPT_MAYBE_CONSTANT_IF (have_sse) { Modes |= mode::xmm128sse; } #endif // MPT_OS #endif // MPT_MODE_KERNEL } } // clang-format on #endif // MPT_ARCH_AMD64 if (Virtualized) { cpuid_result HypervisorVendorID = cpuid(0x4000'0000u); if (HypervisorVendorID.a >= 0x4000'0000u) { HypervisorVendor = HypervisorVendorID.as_text12bcd(); if (HypervisorVendor == std::string_view("Microsoft Hv")) { if (HypervisorVendorID.a >= 0x4000'0001u) { cpuid_result HypervisorInterfaceID = cpuid(0x4000'0001u); HypervisorInterface = HypervisorInterfaceID.as_text4(); } } else if (HypervisorVendor == std::string_view("KVMKVMKVM")) { // nothing } } } } #elif MPT_OS_WINDOWS #if MPT_ARCH_X86 || !MPT_ARCH_X86_AMD64_FAST_DETECT SYSTEM_INFO si = {}; GetSystemInfo(&si); switch (si.wProcessorArchitecture) { case PROCESSOR_ARCHITECTURE_INTEL: case PROCESSOR_ARCHITECTURE_AMD64: case PROCESSOR_ARCHITECTURE_IA32_ON_WIN64: case PROCESSOR_ARCHITECTURE_IA32_ON_ARM64: switch (si.dwProcessorType) { case PROCESSOR_INTEL_386: Family = si.wProcessorLevel; if (((si.wProcessorRevision & 0xff00) >> 8) == 0xff) { Model = ((si.wProcessorRevision & 0x00f0) >> 4) - 0xa; Stepping = ((si.wProcessorRevision & 0x000f) >> 0); } else { Model = ((si.wProcessorRevision & 0xff00) >> 8) + static_cast('A'); Stepping = ((si.wProcessorRevision & 0x00ff) >> 0); } Model = (si.wProcessorRevision & 0xff00) >> 8; Stepping = (si.wProcessorRevision & 0x00ff) >> 0; Features |= featureset::intel386; break; case PROCESSOR_INTEL_486: Family = si.wProcessorLevel; if (((si.wProcessorRevision & 0xff00) >> 8) == 0xff) { Model = ((si.wProcessorRevision & 0x00f0) >> 4) - 0xa; Stepping = ((si.wProcessorRevision & 0x000f) >> 0); } else { Model = ((si.wProcessorRevision & 0xff00) >> 8) + static_cast('A'); Stepping = ((si.wProcessorRevision & 0x00ff) >> 0); } Model = (si.wProcessorRevision & 0xff00) >> 8; Stepping = (si.wProcessorRevision & 0x00ff) >> 0; Features |= featureset::intel486SX; break; case PROCESSOR_INTEL_PENTIUM: Family = si.wProcessorLevel; Model = (si.wProcessorRevision & 0xff00) >> 8; Stepping = (si.wProcessorRevision & 0x00ff) >> 0; // rely on IsProcessorFeaturePresent() for > 486 features // Features |= featureset::intel586; Features |= featureset::intel486SX; break; } break; } Features |= featureset::intel386; const uint8 fpu_level = detect_fpu_level(); if (fpu_level >= 2) { Features |= feature::fpu; } if (fpu_level >= 3) { Features |= feature::fsin; } #elif MPT_ARCH_AMD64 Features |= featureset::amd64; #endif // MPT_ARCH #if MPT_OS_WINDOWS_WINNT // clang-format off Features |= (IsProcessorFeaturePresent(PF_RDTSC_INSTRUCTION_AVAILABLE) != 0) ? (feature::tsc | feature::intel486) : feature::none; Features |= (IsProcessorFeaturePresent(PF_COMPARE_EXCHANGE_DOUBLE) != 0) ? (feature::cx8 | feature::intel486) : feature::none; Features |= (IsProcessorFeaturePresent(PF_MMX_INSTRUCTIONS_AVAILABLE) != 0) ? (feature::mmx | feature::intel486) : feature::none; Features |= (IsProcessorFeaturePresent(PF_3DNOW_INSTRUCTIONS_AVAILABLE) != 0) ? (feature::x3dnow) : feature::none; Features |= (IsProcessorFeaturePresent(PF_XMMI_INSTRUCTIONS_AVAILABLE) != 0) ? (feature::sse | feature::mmxext | feature::fxsr | feature::cmov) : feature::none; Features |= (IsProcessorFeaturePresent(PF_XMMI64_INSTRUCTIONS_AVAILABLE) != 0) ? (feature::sse2) : feature::none; Features |= (IsProcessorFeaturePresent(PF_SSE3_INSTRUCTIONS_AVAILABLE) != 0) ? (feature::sse3) : feature::none; Features |= (IsProcessorFeaturePresent(PF_SSSE3_INSTRUCTIONS_AVAILABLE) != 0) ? (feature::ssse3) : feature::none; Features |= (IsProcessorFeaturePresent(PF_SSE4_1_INSTRUCTIONS_AVAILABLE) != 0) ? (feature::sse4_1) : feature::none; Features |= (IsProcessorFeaturePresent(PF_SSE4_2_INSTRUCTIONS_AVAILABLE) != 0) ? (feature::sse4_1) : feature::none; Features |= (IsProcessorFeaturePresent(PF_AVX_INSTRUCTIONS_AVAILABLE) != 0) ? (feature::avx | feature::xsave) : feature::none; Features |= (IsProcessorFeaturePresent(PF_AVX2_INSTRUCTIONS_AVAILABLE) != 0) ? (feature::avx2 | feature::fma | feature::bmi1) : feature::none; Modes |= (IsProcessorFeaturePresent(PF_XMMI_INSTRUCTIONS_AVAILABLE) != 0) ? (mode::xmm128sse) : mode::base; Modes |= (IsProcessorFeaturePresent(PF_XMMI64_INSTRUCTIONS_AVAILABLE) != 0) ? (mode::xmm128sse) : mode::base; Modes |= (IsProcessorFeaturePresent(PF_SSE3_INSTRUCTIONS_AVAILABLE) != 0) ? (mode::xmm128sse) : mode::base; Modes |= (IsProcessorFeaturePresent(PF_SSSE3_INSTRUCTIONS_AVAILABLE) != 0) ? (mode::xmm128sse) : mode::base; Modes |= (IsProcessorFeaturePresent(PF_SSE4_1_INSTRUCTIONS_AVAILABLE) != 0) ? (mode::xmm128sse) : mode::base; Modes |= (IsProcessorFeaturePresent(PF_SSE4_2_INSTRUCTIONS_AVAILABLE) != 0) ? (mode::xmm128sse) : mode::base; Modes |= (IsProcessorFeaturePresent(PF_AVX_INSTRUCTIONS_AVAILABLE) != 0) ? (mode::xmm128sse | mode::ymm256avx) : mode::base; Modes |= (IsProcessorFeaturePresent(PF_AVX2_INSTRUCTIONS_AVAILABLE) != 0) ? (mode::xmm128sse | mode::ymm256avx) : mode::base; // clang-format on #endif #elif MPT_OS_DJGPP const uint8 cpu_level = detect_cpu_level(); Features |= (cpu_level >= 3) ? featureset::intel386 : feature::none; Features |= (cpu_level >= 4) ? featureset::intel486SX : feature::none; const uint8 fpu_level = detect_fpu_level(); Features |= (fpu_level >= 2) ? feature::fpu : feature::none; Features |= (fpu_level >= 3) ? feature::fsin : feature::none; #endif #endif // MPT_ARCH_X86 || MPT_ARCH_AMD64 } }; struct floating_point { public: static inline constexpr uint16 FCW_IM = (1 << 0); static inline constexpr uint16 FCW_DM = (1 << 1); static inline constexpr uint16 FCW_ZM = (1 << 2); static inline constexpr uint16 FCW_OM = (1 << 3); static inline constexpr uint16 FCW_UM = (1 << 4); static inline constexpr uint16 FCW_PM = (1 << 5); static inline constexpr uint16 FCW_PC = 0x0300; static inline constexpr uint16 FCW_RC = 0x0c00; static inline constexpr uint16 FCW_X = (1 << 12); static inline constexpr uint32 MXCSR_IE = (1 << 0); static inline constexpr uint32 MXCSR_DE = (1 << 1); static inline constexpr uint32 MXCSR_ZE = (1 << 2); static inline constexpr uint32 MXCSR_OE = (1 << 3); static inline constexpr uint32 MXCSR_UE = (1 << 4); static inline constexpr uint32 MXCSR_PE = (1 << 5); static inline constexpr uint32 MXCSR_DAZ = (1 << 6); static inline constexpr uint32 MXCSR_IM = (1 << 7); static inline constexpr uint32 MXCSR_DM = (1 << 8); static inline constexpr uint32 MXCSR_ZM = (1 << 9); static inline constexpr uint32 MXCSR_OM = (1 << 10); static inline constexpr uint32 MXCSR_UM = (1 << 11); static inline constexpr uint32 MXCSR_PM = (1 << 12); static inline constexpr uint32 MXCSR_RC = (1 << 13) | (1 << 14); static inline constexpr uint32 MXCSR_FTZ = (1 << 15); enum class precision : uint8 { single24 = 0, reserved = 1, double53 = 2, extended64 = 3, }; enum class rounding : uint8 { nearest = 0, down = 1, up = 2, zero = 3, }; struct alignas(16) fxsave_state { uint16 fcw; uint16 fsw; uint16 ftw; uint16 fop; uint32 fip; uint32 fcs; uint32 foo; uint32 fos; uint32 mxcsr; uint32 mxcsr_mask; uint32 st_space[32]; uint32 xmm_space[32]; uint8 padding[224]; }; static_assert(sizeof(fxsave_state) == 512); struct control_state { uint8 x87_level = 0; uint16 x87fcw = 0; // default 0x37f (glibc) / 0x27f (msvc) uint32 mxcsr_mask = 0; uint32 mxcsr = 0; // default: 0x00001f80 }; #if MPT_ARCH_X86 #if MPT_COMPILER_MSVC [[nodiscard]] static MPT_FORCEINLINE uint16 get_x87fcw() noexcept { uint16 tmp = 0; // clang-format off _asm { fwait fnstcw tmp } // clang-format on return tmp; } static MPT_FORCEINLINE void set_x87fcw(uint16 fcw) noexcept { // clang-format off _asm { fldcw fcw } // clang-format on } [[nodiscard]] static MPT_FORCEINLINE uint32 get_mxcsr() noexcept { return _mm_getcsr(); } static MPT_FORCEINLINE void set_mxcsr(uint32 csr) noexcept { _mm_setcsr(csr); } static MPT_FORCEINLINE void fxsave(fxsave_state * state) noexcept { _fxsave(state); } static MPT_FORCEINLINE void fxrstor(const fxsave_state * state) noexcept { _fxrstor(state); } #elif MPT_COMPILER_GCC || MPT_COMPILER_CLANG [[nodiscard]] static MPT_FORCEINLINE uint16 get_x87fcw() noexcept { typedef unsigned int fpu_control_t __attribute__((__mode__(__HI__))); fpu_control_t tmp = 0; // clang-format off __asm__ __volatile__("fwait" "\n" "fnstcw %0" : "=m" (*&tmp)); // clang-format on return static_cast(tmp); } static MPT_FORCEINLINE void set_x87fcw(uint16 fcw) noexcept { typedef unsigned int fpu_control_t __attribute__((__mode__(__HI__))); fpu_control_t tmp = fcw; // clang-format off __asm__ __volatile__("fldcw %0" : : "m" (*&tmp)); // clang-format on } [[nodiscard]] static MPT_FORCEINLINE uint32 get_mxcsr() noexcept { #ifdef MPT_ARCH_X86_SSE return __builtin_ia32_stmxcsr(); #else uint32 csr = 0; // clang-format off __asm__ __volatile__("stmxcsr %0" : "=m" (csr)); // clang-format on return csr; #endif } static MPT_FORCEINLINE void set_mxcsr(uint32 csr) noexcept { #ifdef MPT_ARCH_X86_SSE #if MPT_COMPILER_GCC // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55752 std::atomic_thread_fence(std::memory_order_seq_cst); #endif __builtin_ia32_ldmxcsr(csr); #if MPT_COMPILER_GCC // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55752 std::atomic_thread_fence(std::memory_order_seq_cst); #endif #else // clang-format off __asm__ __volatile__("ldmxcsr %0" : : "m" (csr)); // clang-format on #endif } static MPT_FORCEINLINE void fxsave(fxsave_state * state) noexcept { #ifdef MPT_ARCH_X86_FXSR __builtin_ia32_fxsave(state); #else // clang-format off __asm__ __volatile__("fxsave %0" : : "m" (*state)); // clang-format on #endif } static MPT_FORCEINLINE void fxrstor(const fxsave_state * state) noexcept { #ifdef MPT_ARCH_X86_FXSR __builtin_ia32_fxrstor(const_cast(state)); #else // clang-format off __asm__ __volatile__("fxrstor %0" : : "m" (*state)); // clang-format on #endif } #endif // MPT_COMPILER static MPT_FORCEINLINE bool have_fxsr() noexcept { #ifdef MPT_ARCH_X86_FXSR return true; #else return cpu_info{}[mpt::arch::x86::feature::fxsr]; #endif } static MPT_FORCEINLINE bool have_sse() noexcept { #ifdef MPT_ARCH_X86_SSE return true; #else const cpu_info cpu_info; return cpu_info[mpt::arch::x86::feature::sse] && cpu_info[mpt::arch::x86::mode::xmm128sse]; #endif } static MPT_FORCEINLINE uint8 get_fpu_level() noexcept { #ifdef MPT_ARCH_X86_FSIN return 3; #elif defined(MPT_ARCH_X86_FPU) return cpu_info{}[mpt::arch::x86::feature::fsin] ? 3 : 2; #else cpu_info tmp{}; return tmp[mpt::arch::x86::feature::fsin] ? 3 : tmp[mpt::arch::x86::feature::fpu] ? 2 : 0; #endif } static MPT_FORCEINLINE control_state get_state() noexcept { control_state result; #ifdef MPT_ARCH_X86_FXSR fxsave_state tmp = {}; fxsave(&tmp); result.x87_level = 3; result.x87fcw = tmp.fcw; result.mxcsr_mask = tmp.mxcsr_mask; result.mxcsr = tmp.mxcsr; #else if (have_fxsr()) { fxsave_state tmp = {}; fxsave(&tmp); result.x87_level = 3; result.x87fcw = tmp.fcw; result.mxcsr_mask = tmp.mxcsr_mask; result.mxcsr = tmp.mxcsr; } else { result.x87_level = get_fpu_level(); if (result.x87_level > 0) { result.x87fcw = get_x87fcw(); } } #endif return result; } static MPT_FORCEINLINE void set_state(control_state state) noexcept { #ifdef MPT_ARCH_X86_SSE if (state.x87_level) { set_x87fcw(state.x87fcw); } if (state.mxcsr_mask) { set_mxcsr(state.mxcsr); } #else if (have_sse()) { if (state.x87_level) { set_x87fcw(state.x87fcw); } if (state.mxcsr_mask) { set_mxcsr(state.mxcsr); } } else { #ifdef MPT_ARCH_X86_FXSR fxsave_state tmp = {}; fxsave(&tmp); if (state.x87_level) { tmp.fcw = state.x87fcw; } if (state.mxcsr_mask) { tmp.mxcsr = state.mxcsr; } fxrstor(&tmp); #else if (have_fxsr()) { fxsave_state tmp = {}; fxsave(&tmp); if (state.x87_level) { tmp.fcw = state.x87fcw; } if (state.mxcsr_mask) { tmp.mxcsr = state.mxcsr; } fxrstor(&tmp); } else { if (state.x87_level) { set_x87fcw(state.x87fcw); } } #endif } #endif } #elif MPT_ARCH_AMD64 #if MPT_COMPILER_MSVC [[nodiscard]] static MPT_FORCEINLINE uint16 get_x87fcw() noexcept { fxsave_state state = {}; fxsave(&state); return state.fcw; } static MPT_FORCEINLINE void set_x87fcw(uint16 fcw) noexcept { fxsave_state state = {}; fxsave(&state); state.fcw = fcw; fxrstor(&state); } [[nodiscard]] static MPT_FORCEINLINE uint32 get_mxcsr() noexcept { return _mm_getcsr(); } static MPT_FORCEINLINE void set_mxcsr(uint32 csr) noexcept { _mm_setcsr(csr); } static MPT_FORCEINLINE void fxsave(fxsave_state * state) noexcept { _fxsave(state); } static MPT_FORCEINLINE void fxrstor(const fxsave_state * state) noexcept { _fxrstor(state); } static MPT_FORCEINLINE bool have_fxsr() noexcept { return true; } static MPT_FORCEINLINE control_state get_state() noexcept { control_state result; fxsave_state tmp = {}; fxsave(&tmp); result.x87_level = 3; result.x87fcw = tmp.fcw; result.mxcsr_mask = tmp.mxcsr_mask; result.mxcsr = tmp.mxcsr; return result; } static MPT_FORCEINLINE void set_state(control_state state) noexcept { fxsave_state tmp = {}; fxsave(&tmp); tmp.fcw = state.x87fcw; tmp.mxcsr_mask = state.mxcsr_mask; tmp.mxcsr = state.mxcsr; fxrstor(&tmp); } #elif MPT_COMPILER_GCC || MPT_COMPILER_CLANG [[nodiscard]] static MPT_FORCEINLINE uint16 get_x87fcw() noexcept { typedef unsigned int fpu_control_t __attribute__((__mode__(__HI__))); fpu_control_t tmp = 0; // clang-format off __asm__ __volatile__("fwait" "\n" "fnstcw %0" : "=m" (*&tmp)); // clang-format on return static_cast(tmp); } static MPT_FORCEINLINE void set_x87fcw(uint16 fcw) noexcept { typedef unsigned int fpu_control_t __attribute__((__mode__(__HI__))); fpu_control_t tmp = fcw; // clang-format off __asm__ __volatile__("fldcw %0" : : "m" (*&tmp)); // clang-format on } [[nodiscard]] static MPT_FORCEINLINE uint32 get_mxcsr() noexcept { return __builtin_ia32_stmxcsr(); } static MPT_FORCEINLINE void set_mxcsr(uint32 csr) noexcept { #if MPT_COMPILER_GCC // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55752 std::atomic_thread_fence(std::memory_order_seq_cst); #endif __builtin_ia32_ldmxcsr(csr); #if MPT_COMPILER_GCC // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55752 std::atomic_thread_fence(std::memory_order_seq_cst); #endif } static MPT_FORCEINLINE void fxsave(fxsave_state * state) noexcept { __builtin_ia32_fxsave(state); } static MPT_FORCEINLINE void fxrstor(const fxsave_state * state) noexcept { __builtin_ia32_fxrstor(const_cast(state)); } static MPT_FORCEINLINE bool have_fxsr() noexcept { return true; } static MPT_FORCEINLINE control_state get_state() noexcept { control_state result; result.x87_level = 3; result.x87fcw = get_x87fcw(); result.mxcsr_mask = 0x0000'ffff; result.mxcsr = get_mxcsr(); return result; } static MPT_FORCEINLINE void set_state(control_state state) noexcept { set_x87fcw(state.x87fcw); set_mxcsr(state.mxcsr); } #endif // MPT_COMPILER class guard { private: const control_state m_oldstate; public: MPT_FORCEINLINE guard(std::optional rounding, std::optional denormals_as_zero, std::optional precision, std::optional infinity_projective) noexcept : m_oldstate(get_state()) { control_state state = m_oldstate; if (rounding) { if (state.x87_level) { state.x87fcw = (state.x87fcw & ~FCW_RC) | (mpt::to_underlying(*rounding) << mpt::countr_zero(FCW_RC)); } if ((state.mxcsr_mask & MXCSR_RC) == MXCSR_RC) { state.mxcsr = (state.mxcsr & ~MXCSR_RC) | (mpt::to_underlying(*rounding) << mpt::countr_zero(MXCSR_RC)); } } if (denormals_as_zero) { if (state.mxcsr_mask) { state.mxcsr = (state.mxcsr & ~(MXCSR_FTZ | MXCSR_DAZ)) | ((*denormals_as_zero) ? ((MXCSR_FTZ | MXCSR_DAZ) & state.mxcsr_mask) : 0); } } if (precision) { if (state.x87_level) { state.x87fcw = (state.x87fcw & ~FCW_PC) | (mpt::to_underlying(*precision) << mpt::countr_zero(FCW_PC)); } } if (infinity_projective) { if (state.x87_level <= 2) { state.x87fcw = (state.x87fcw & ~FCW_X) | ((*infinity_projective) ? 0 : FCW_X); } } set_state(state); } MPT_FORCEINLINE ~guard() { set_state(m_oldstate); } }; #endif // MPT_ARCH }; } // namespace x86 namespace amd64 = x86; } // namespace arch } // namespace MPT_INLINE_NS } // namespace mpt #endif // MPT_ARCH_X86_AMD64_HPP