From a1942238f5046811833e9636c64f991905960732 Mon Sep 17 00:00:00 2001 From: bunnei Date: Wed, 12 Aug 2015 00:19:20 -0400 Subject: Common: Move cpu_detect to x64 directory. --- src/common/x64/cpu_detect.cpp | 229 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 229 insertions(+) create mode 100644 src/common/x64/cpu_detect.cpp (limited to 'src/common/x64/cpu_detect.cpp') diff --git a/src/common/x64/cpu_detect.cpp b/src/common/x64/cpu_detect.cpp new file mode 100644 index 000000000..72c8297a3 --- /dev/null +++ b/src/common/x64/cpu_detect.cpp @@ -0,0 +1,229 @@ +// Copyright 2008 Dolphin Emulator Project +// Licensed under GPLv2+ +// Refer to the license.txt file included. + +#include +#include + +#include "common/common_types.h" + +#include "cpu_detect.h" + +#ifndef _WIN32 + +#ifdef __FreeBSD__ +#include +#include +#endif + +static inline void __cpuidex(int info[4], int function_id, int subfunction_id) +{ +#ifdef __FreeBSD__ + // Despite the name, this is just do_cpuid() with ECX as second input. + cpuid_count((u_int)function_id, (u_int)subfunction_id, (u_int*)info); +#else + info[0] = function_id; // eax + info[2] = subfunction_id; // ecx + __asm__( + "cpuid" + : "=a" (info[0]), + "=b" (info[1]), + "=c" (info[2]), + "=d" (info[3]) + : "a" (function_id), + "c" (subfunction_id) + ); +#endif +} + +static inline void __cpuid(int info[4], int function_id) +{ + return __cpuidex(info, function_id, 0); +} + +#define _XCR_XFEATURE_ENABLED_MASK 0 +static u64 _xgetbv(u32 index) +{ + u32 eax, edx; + __asm__ __volatile__("xgetbv" : "=a"(eax), "=d"(edx) : "c"(index)); + return ((u64)edx << 32) | eax; +} + +#endif // ifndef _WIN32 + +namespace Common { + +CPUInfo cpu_info; + +CPUInfo::CPUInfo() { + Detect(); +} + +// Detects the various CPU features +void CPUInfo::Detect() { + memset(this, 0, sizeof(*this)); +#ifdef ARCHITECTURE_X64 + Mode64bit = true; + OS64bit = true; +#endif + num_cores = 1; + + // Set obvious defaults, for extra safety + if (Mode64bit) { + bSSE = true; + bSSE2 = true; + bLongMode = true; + } + + // Assume CPU supports the CPUID instruction. Those that don't can barely + // boot modern OS:es anyway. + int cpu_id[4]; + memset(brand_string, 0, sizeof(brand_string)); + + // Detect CPU's CPUID capabilities, and grab CPU string + __cpuid(cpu_id, 0x00000000); + u32 max_std_fn = cpu_id[0]; // EAX + *((int *)brand_string) = cpu_id[1]; + *((int *)(brand_string + 4)) = cpu_id[3]; + *((int *)(brand_string + 8)) = cpu_id[2]; + __cpuid(cpu_id, 0x80000000); + u32 max_ex_fn = cpu_id[0]; + if (!strcmp(brand_string, "GenuineIntel")) + vendor = VENDOR_INTEL; + else if (!strcmp(brand_string, "AuthenticAMD")) + vendor = VENDOR_AMD; + else + vendor = VENDOR_OTHER; + + // Set reasonable default brand string even if brand string not available. + strcpy(cpu_string, brand_string); + + // Detect family and other misc stuff. + bool ht = false; + HTT = ht; + logical_cpu_count = 1; + if (max_std_fn >= 1) { + __cpuid(cpu_id, 0x00000001); + int family = ((cpu_id[0] >> 8) & 0xf) + ((cpu_id[0] >> 20) & 0xff); + int model = ((cpu_id[0] >> 4) & 0xf) + ((cpu_id[0] >> 12) & 0xf0); + // Detect people unfortunate enough to be running Dolphin on an Atom + if (family == 6 && (model == 0x1C || model == 0x26 || model == 0x27 || model == 0x35 || model == 0x36 || + model == 0x37 || model == 0x4A || model == 0x4D || model == 0x5A || model == 0x5D)) + bAtom = true; + logical_cpu_count = (cpu_id[1] >> 16) & 0xFF; + ht = (cpu_id[3] >> 28) & 1; + + if ((cpu_id[3] >> 25) & 1) bSSE = true; + if ((cpu_id[3] >> 26) & 1) bSSE2 = true; + if ((cpu_id[2]) & 1) bSSE3 = true; + if ((cpu_id[2] >> 9) & 1) bSSSE3 = true; + if ((cpu_id[2] >> 19) & 1) bSSE4_1 = true; + if ((cpu_id[2] >> 20) & 1) bSSE4_2 = true; + if ((cpu_id[2] >> 22) & 1) bMOVBE = true; + if ((cpu_id[2] >> 25) & 1) bAES = true; + + if ((cpu_id[3] >> 24) & 1) + { + // We can use FXSAVE. + bFXSR = true; + } + + // AVX support requires 3 separate checks: + // - Is the AVX bit set in CPUID? + // - Is the XSAVE bit set in CPUID? + // - XGETBV result has the XCR bit set. + if (((cpu_id[2] >> 28) & 1) && ((cpu_id[2] >> 27) & 1)) { + if ((_xgetbv(_XCR_XFEATURE_ENABLED_MASK) & 0x6) == 0x6) { + bAVX = true; + if ((cpu_id[2] >> 12) & 1) + bFMA = true; + } + } + + if (max_std_fn >= 7) { + __cpuidex(cpu_id, 0x00000007, 0x00000000); + // careful; we can't enable AVX2 unless the XSAVE/XGETBV checks above passed + if ((cpu_id[1] >> 5) & 1) + bAVX2 = bAVX; + if ((cpu_id[1] >> 3) & 1) + bBMI1 = true; + if ((cpu_id[1] >> 8) & 1) + bBMI2 = true; + } + } + + bFlushToZero = bSSE; + + if (max_ex_fn >= 0x80000004) { + // Extract CPU model string + __cpuid(cpu_id, 0x80000002); + memcpy(cpu_string, cpu_id, sizeof(cpu_id)); + __cpuid(cpu_id, 0x80000003); + memcpy(cpu_string + 16, cpu_id, sizeof(cpu_id)); + __cpuid(cpu_id, 0x80000004); + memcpy(cpu_string + 32, cpu_id, sizeof(cpu_id)); + } + if (max_ex_fn >= 0x80000001) { + // Check for more features. + __cpuid(cpu_id, 0x80000001); + if (cpu_id[2] & 1) bLAHFSAHF64 = true; + if ((cpu_id[2] >> 5) & 1) bLZCNT = true; + if ((cpu_id[2] >> 16) & 1) bFMA4 = true; + if ((cpu_id[3] >> 29) & 1) bLongMode = true; + } + + num_cores = (logical_cpu_count == 0) ? 1 : logical_cpu_count; + + if (max_ex_fn >= 0x80000008) { + // Get number of cores. This is a bit complicated. Following AMD manual here. + __cpuid(cpu_id, 0x80000008); + int apic_id_core_id_size = (cpu_id[2] >> 12) & 0xF; + if (apic_id_core_id_size == 0) { + if (ht) { + // New mechanism for modern Intel CPUs. + if (vendor == VENDOR_INTEL) { + __cpuidex(cpu_id, 0x00000004, 0x00000000); + int cores_x_package = ((cpu_id[0] >> 26) & 0x3F) + 1; + HTT = (cores_x_package < logical_cpu_count); + cores_x_package = ((logical_cpu_count % cores_x_package) == 0) ? cores_x_package : 1; + num_cores = (cores_x_package > 1) ? cores_x_package : num_cores; + logical_cpu_count /= cores_x_package; + } + } + } else { + // Use AMD's new method. + num_cores = (cpu_id[2] & 0xFF) + 1; + } + } +} + +// Turn the CPU info into a string we can show +std::string CPUInfo::Summarize() { + std::string sum(cpu_string); + sum += " ("; + sum += brand_string; + sum += ")"; + + if (bSSE) sum += ", SSE"; + if (bSSE2) { + sum += ", SSE2"; + if (!bFlushToZero) + sum += " (but not DAZ!)"; + } + if (bSSE3) sum += ", SSE3"; + if (bSSSE3) sum += ", SSSE3"; + if (bSSE4_1) sum += ", SSE4.1"; + if (bSSE4_2) sum += ", SSE4.2"; + if (HTT) sum += ", HTT"; + if (bAVX) sum += ", AVX"; + if (bAVX2) sum += ", AVX2"; + if (bBMI1) sum += ", BMI1"; + if (bBMI2) sum += ", BMI2"; + if (bFMA) sum += ", FMA"; + if (bAES) sum += ", AES"; + if (bMOVBE) sum += ", MOVBE"; + if (bLongMode) sum += ", 64-bit support"; + return sum; +} + +} // namespace Common -- cgit v1.2.3 From 0ee00861f6747f8946972a91539e857f493e9cc6 Mon Sep 17 00:00:00 2001 From: bunnei Date: Wed, 12 Aug 2015 17:42:13 -0400 Subject: Common: Cleanup CPU capability detection code. --- src/common/x64/cpu_detect.cpp | 220 +++++++++++++++++------------------------- 1 file changed, 89 insertions(+), 131 deletions(-) (limited to 'src/common/x64/cpu_detect.cpp') diff --git a/src/common/x64/cpu_detect.cpp b/src/common/x64/cpu_detect.cpp index 72c8297a3..d9c430c67 100644 --- a/src/common/x64/cpu_detect.cpp +++ b/src/common/x64/cpu_detect.cpp @@ -1,23 +1,25 @@ -// Copyright 2008 Dolphin Emulator Project -// Licensed under GPLv2+ +// Copyright 2013 Dolphin Emulator Project / 2015 Citra Emulator Project +// Licensed under GPLv2 or any later version // Refer to the license.txt file included. #include #include +#include #include "common/common_types.h" #include "cpu_detect.h" -#ifndef _WIN32 +namespace Common { + +#ifndef _MSC_VER #ifdef __FreeBSD__ #include #include #endif -static inline void __cpuidex(int info[4], int function_id, int subfunction_id) -{ +static inline void __cpuidex(int info[4], int function_id, int subfunction_id) { #ifdef __FreeBSD__ // Despite the name, this is just do_cpuid() with ECX as second input. cpuid_count((u_int)function_id, (u_int)subfunction_id, (u_int*)info); @@ -36,96 +38,67 @@ static inline void __cpuidex(int info[4], int function_id, int subfunction_id) #endif } -static inline void __cpuid(int info[4], int function_id) -{ +static inline void __cpuid(int info[4], int function_id) { return __cpuidex(info, function_id, 0); } #define _XCR_XFEATURE_ENABLED_MASK 0 -static u64 _xgetbv(u32 index) -{ +static u64 _xgetbv(u32 index) { u32 eax, edx; __asm__ __volatile__("xgetbv" : "=a"(eax), "=d"(edx) : "c"(index)); return ((u64)edx << 32) | eax; } -#endif // ifndef _WIN32 - -namespace Common { - -CPUInfo cpu_info; - -CPUInfo::CPUInfo() { - Detect(); -} +#endif // ifndef _MSC_VER // Detects the various CPU features -void CPUInfo::Detect() { - memset(this, 0, sizeof(*this)); -#ifdef ARCHITECTURE_X64 - Mode64bit = true; - OS64bit = true; -#endif - num_cores = 1; +static CPUCaps Detect() { + CPUCaps caps = {}; - // Set obvious defaults, for extra safety - if (Mode64bit) { - bSSE = true; - bSSE2 = true; - bLongMode = true; - } + caps.num_cores = std::thread::hardware_concurrency(); + + // Assumes the CPU supports the CPUID instruction. Those that don't would likely not support + // Citra at all anyway - // Assume CPU supports the CPUID instruction. Those that don't can barely - // boot modern OS:es anyway. int cpu_id[4]; - memset(brand_string, 0, sizeof(brand_string)); + memset(caps.brand_string, 0, sizeof(caps.brand_string)); - // Detect CPU's CPUID capabilities, and grab CPU string + // Detect CPU's CPUID capabilities and grab CPU string __cpuid(cpu_id, 0x00000000); - u32 max_std_fn = cpu_id[0]; // EAX - *((int *)brand_string) = cpu_id[1]; - *((int *)(brand_string + 4)) = cpu_id[3]; - *((int *)(brand_string + 8)) = cpu_id[2]; + u32 max_std_fn = cpu_id[0]; // EAX + + std::memcpy(&caps.brand_string[0], &cpu_id[1], sizeof(int)); + std::memcpy(&caps.brand_string[4], &cpu_id[3], sizeof(int)); + std::memcpy(&caps.brand_string[8], &cpu_id[2], sizeof(int)); + __cpuid(cpu_id, 0x80000000); + u32 max_ex_fn = cpu_id[0]; - if (!strcmp(brand_string, "GenuineIntel")) - vendor = VENDOR_INTEL; - else if (!strcmp(brand_string, "AuthenticAMD")) - vendor = VENDOR_AMD; + if (!strcmp(caps.brand_string, "GenuineIntel")) + caps.vendor = CPUVendor::INTEL; + else if (!strcmp(caps.brand_string, "AuthenticAMD")) + caps.vendor = CPUVendor::AMD; else - vendor = VENDOR_OTHER; + caps.vendor = CPUVendor::OTHER; - // Set reasonable default brand string even if brand string not available. - strcpy(cpu_string, brand_string); + // Set reasonable default brand string even if brand string not available + strcpy(caps.cpu_string, caps.brand_string); - // Detect family and other misc stuff. - bool ht = false; - HTT = ht; - logical_cpu_count = 1; + // Detect family and other miscellaneous features if (max_std_fn >= 1) { __cpuid(cpu_id, 0x00000001); - int family = ((cpu_id[0] >> 8) & 0xf) + ((cpu_id[0] >> 20) & 0xff); - int model = ((cpu_id[0] >> 4) & 0xf) + ((cpu_id[0] >> 12) & 0xf0); - // Detect people unfortunate enough to be running Dolphin on an Atom - if (family == 6 && (model == 0x1C || model == 0x26 || model == 0x27 || model == 0x35 || model == 0x36 || - model == 0x37 || model == 0x4A || model == 0x4D || model == 0x5A || model == 0x5D)) - bAtom = true; - logical_cpu_count = (cpu_id[1] >> 16) & 0xFF; - ht = (cpu_id[3] >> 28) & 1; - - if ((cpu_id[3] >> 25) & 1) bSSE = true; - if ((cpu_id[3] >> 26) & 1) bSSE2 = true; - if ((cpu_id[2]) & 1) bSSE3 = true; - if ((cpu_id[2] >> 9) & 1) bSSSE3 = true; - if ((cpu_id[2] >> 19) & 1) bSSE4_1 = true; - if ((cpu_id[2] >> 20) & 1) bSSE4_2 = true; - if ((cpu_id[2] >> 22) & 1) bMOVBE = true; - if ((cpu_id[2] >> 25) & 1) bAES = true; - - if ((cpu_id[3] >> 24) & 1) - { - // We can use FXSAVE. - bFXSR = true; + + if ((cpu_id[3] >> 25) & 1) caps.sse = true; + if ((cpu_id[3] >> 26) & 1) caps.sse2 = true; + if ((cpu_id[2]) & 1) caps.sse3 = true; + if ((cpu_id[2] >> 9) & 1) caps.ssse3 = true; + if ((cpu_id[2] >> 19) & 1) caps.sse4_1 = true; + if ((cpu_id[2] >> 20) & 1) caps.sse4_2 = true; + if ((cpu_id[2] >> 22) & 1) caps.movbe = true; + if ((cpu_id[2] >> 25) & 1) caps.aes = true; + + if ((cpu_id[3] >> 24) & 1) { + caps.fxsave_fxrstor = true; } // AVX support requires 3 separate checks: @@ -134,95 +107,80 @@ void CPUInfo::Detect() { // - XGETBV result has the XCR bit set. if (((cpu_id[2] >> 28) & 1) && ((cpu_id[2] >> 27) & 1)) { if ((_xgetbv(_XCR_XFEATURE_ENABLED_MASK) & 0x6) == 0x6) { - bAVX = true; + caps.avx = true; if ((cpu_id[2] >> 12) & 1) - bFMA = true; + caps.fma = true; } } if (max_std_fn >= 7) { __cpuidex(cpu_id, 0x00000007, 0x00000000); - // careful; we can't enable AVX2 unless the XSAVE/XGETBV checks above passed + // Can't enable AVX2 unless the XSAVE/XGETBV checks above passed if ((cpu_id[1] >> 5) & 1) - bAVX2 = bAVX; + caps.avx2 = caps.avx; if ((cpu_id[1] >> 3) & 1) - bBMI1 = true; + caps.bmi1 = true; if ((cpu_id[1] >> 8) & 1) - bBMI2 = true; + caps.bmi2 = true; } } - bFlushToZero = bSSE; + caps.flush_to_zero = caps.sse; if (max_ex_fn >= 0x80000004) { // Extract CPU model string __cpuid(cpu_id, 0x80000002); - memcpy(cpu_string, cpu_id, sizeof(cpu_id)); + std::memcpy(caps.cpu_string, cpu_id, sizeof(cpu_id)); __cpuid(cpu_id, 0x80000003); - memcpy(cpu_string + 16, cpu_id, sizeof(cpu_id)); + std::memcpy(caps.cpu_string + 16, cpu_id, sizeof(cpu_id)); __cpuid(cpu_id, 0x80000004); - memcpy(cpu_string + 32, cpu_id, sizeof(cpu_id)); + std::memcpy(caps.cpu_string + 32, cpu_id, sizeof(cpu_id)); } + if (max_ex_fn >= 0x80000001) { - // Check for more features. + // Check for more features __cpuid(cpu_id, 0x80000001); - if (cpu_id[2] & 1) bLAHFSAHF64 = true; - if ((cpu_id[2] >> 5) & 1) bLZCNT = true; - if ((cpu_id[2] >> 16) & 1) bFMA4 = true; - if ((cpu_id[3] >> 29) & 1) bLongMode = true; + if (cpu_id[2] & 1) caps.lahf_sahf_64 = true; + if ((cpu_id[2] >> 5) & 1) caps.lzcnt = true; + if ((cpu_id[2] >> 16) & 1) caps.fma4 = true; + if ((cpu_id[3] >> 29) & 1) caps.long_mode = true; } - num_cores = (logical_cpu_count == 0) ? 1 : logical_cpu_count; - - if (max_ex_fn >= 0x80000008) { - // Get number of cores. This is a bit complicated. Following AMD manual here. - __cpuid(cpu_id, 0x80000008); - int apic_id_core_id_size = (cpu_id[2] >> 12) & 0xF; - if (apic_id_core_id_size == 0) { - if (ht) { - // New mechanism for modern Intel CPUs. - if (vendor == VENDOR_INTEL) { - __cpuidex(cpu_id, 0x00000004, 0x00000000); - int cores_x_package = ((cpu_id[0] >> 26) & 0x3F) + 1; - HTT = (cores_x_package < logical_cpu_count); - cores_x_package = ((logical_cpu_count % cores_x_package) == 0) ? cores_x_package : 1; - num_cores = (cores_x_package > 1) ? cores_x_package : num_cores; - logical_cpu_count /= cores_x_package; - } - } - } else { - // Use AMD's new method. - num_cores = (cpu_id[2] & 0xFF) + 1; - } - } + return caps; } -// Turn the CPU info into a string we can show -std::string CPUInfo::Summarize() { - std::string sum(cpu_string); +const CPUCaps& GetCPUCaps() { + static CPUCaps caps = Detect(); + return caps; +} + +std::string GetCPUCapsString() { + auto caps = GetCPUCaps(); + + std::string sum(caps.cpu_string); sum += " ("; - sum += brand_string; + sum += caps.brand_string; sum += ")"; - if (bSSE) sum += ", SSE"; - if (bSSE2) { + if (caps.sse) sum += ", SSE"; + if (caps.sse2) { sum += ", SSE2"; - if (!bFlushToZero) - sum += " (but not DAZ!)"; + if (!caps.flush_to_zero) sum += " (without DAZ)"; } - if (bSSE3) sum += ", SSE3"; - if (bSSSE3) sum += ", SSSE3"; - if (bSSE4_1) sum += ", SSE4.1"; - if (bSSE4_2) sum += ", SSE4.2"; - if (HTT) sum += ", HTT"; - if (bAVX) sum += ", AVX"; - if (bAVX2) sum += ", AVX2"; - if (bBMI1) sum += ", BMI1"; - if (bBMI2) sum += ", BMI2"; - if (bFMA) sum += ", FMA"; - if (bAES) sum += ", AES"; - if (bMOVBE) sum += ", MOVBE"; - if (bLongMode) sum += ", 64-bit support"; + + if (caps.sse3) sum += ", SSE3"; + if (caps.ssse3) sum += ", SSSE3"; + if (caps.sse4_1) sum += ", SSE4.1"; + if (caps.sse4_2) sum += ", SSE4.2"; + if (caps.avx) sum += ", AVX"; + if (caps.avx2) sum += ", AVX2"; + if (caps.bmi1) sum += ", BMI1"; + if (caps.bmi2) sum += ", BMI2"; + if (caps.fma) sum += ", FMA"; + if (caps.aes) sum += ", AES"; + if (caps.movbe) sum += ", MOVBE"; + if (caps.long_mode) sum += ", 64-bit support"; + return sum; } -- cgit v1.2.3