diff options
| author | Jasper St. Pierre <jstpierre@mecheye.net> | 2013-12-07 15:14:29 -0500 |
|---|---|---|
| committer | Jasper St. Pierre <jstpierre@mecheye.net> | 2013-12-31 14:03:19 -0500 |
| commit | 34692ab826abc8f8faa61bdb2280b742424528f1 (patch) | |
| tree | b0eedc645badbea316ff59c125eaf808b28777ec /Source/Core/Common/x64CPUDetect.cpp | |
| parent | 43e618682e8093602fc83dfa902ee7ae3fd3a4f6 (diff) | |
Remove unnecessary Src/ folders
Diffstat (limited to 'Source/Core/Common/x64CPUDetect.cpp')
| -rw-r--r-- | Source/Core/Common/x64CPUDetect.cpp | 274 |
1 files changed, 274 insertions, 0 deletions
diff --git a/Source/Core/Common/x64CPUDetect.cpp b/Source/Core/Common/x64CPUDetect.cpp new file mode 100644 index 0000000000..b9ed2b538a --- /dev/null +++ b/Source/Core/Common/x64CPUDetect.cpp @@ -0,0 +1,274 @@ +// Copyright 2013 Dolphin Emulator Project +// Licensed under GPLv2 +// Refer to the license.txt file included. + +#include <memory.h> +#include "Common.h" + +#ifdef _WIN32 +#define _interlockedbittestandset workaround_ms_header_bug_platform_sdk6_set +#define _interlockedbittestandreset workaround_ms_header_bug_platform_sdk6_reset +#define _interlockedbittestandset64 workaround_ms_header_bug_platform_sdk6_set64 +#define _interlockedbittestandreset64 workaround_ms_header_bug_platform_sdk6_reset64 +#include <intrin.h> +#undef _interlockedbittestandset +#undef _interlockedbittestandreset +#undef _interlockedbittestandset64 +#undef _interlockedbittestandreset64 +#else + +//#include <config/i386/cpuid.h> +#ifndef _M_GENERIC +#include <xmmintrin.h> +#endif + +#if defined __FreeBSD__ +#include <sys/types.h> +#include <machine/cpufunc.h> +#else +static inline void do_cpuid(unsigned int *eax, unsigned int *ebx, + unsigned int *ecx, unsigned int *edx) +{ +#if defined _M_GENERIC + (*eax) = (*ebx) = (*ecx) = (*edx) = 0; +#elif defined _LP64 + // Note: EBX is reserved on Mac OS X and in PIC on Linux, so it has to + // restored at the end of the asm block. + __asm__ ( + "cpuid;" + "movl %%ebx,%1;" + : "=a" (*eax), + "=S" (*ebx), + "=c" (*ecx), + "=d" (*edx) + : "a" (*eax) + : "rbx" + ); +#else + __asm__ ( + "cpuid;" + "movl %%ebx,%1;" + : "=a" (*eax), + "=S" (*ebx), + "=c" (*ecx), + "=d" (*edx) + : "a" (*eax) + : "ebx" + ); +#endif +} +#endif /* defined __FreeBSD__ */ + +static void __cpuid(int info[4], int x) +{ +#if defined __FreeBSD__ + do_cpuid((unsigned int)x, (unsigned int*)info); +#else + unsigned int eax = x, ebx = 0, ecx = 0, edx = 0; + do_cpuid(&eax, &ebx, &ecx, &edx); + info[0] = eax; + info[1] = ebx; + info[2] = ecx; + info[3] = edx; +#endif +} + +#define _XCR_XFEATURE_ENABLED_MASK 0 +static unsigned long long _xgetbv(unsigned int index) +{ +#ifndef _M_GENERIC + unsigned int eax, edx; + __asm__ __volatile__("xgetbv" : "=a"(eax), "=d"(edx) : "c"(index)); + return ((unsigned long long)edx << 32) | eax; +#endif +} + +#endif + +#include "Common.h" +#include "CPUDetect.h" +#include "StringUtil.h" + +CPUInfo cpu_info; + +CPUInfo::CPUInfo() { + Detect(); +} + +// Detects the various cpu features +void CPUInfo::Detect() +{ + memset(this, 0, sizeof(*this)); +#ifdef _M_IX86 + Mode64bit = false; +#elif defined (_M_X64) + Mode64bit = true; + OS64bit = true; +#endif + num_cores = 1; + +#ifdef _WIN32 +#ifdef _M_IX86 + BOOL f64 = false; + IsWow64Process(GetCurrentProcess(), &f64); + OS64bit = (f64 == TRUE) ? true : false; +#endif +#endif + + // 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(cpu_string, 0, sizeof(cpu_string)); + + // Detect CPU's CPUID capabilities, and grab cpu string + __cpuid(cpu_id, 0x00000000); + u32 max_std_fn = cpu_id[0]; // EAX + *((int *)cpu_string) = cpu_id[1]; + *((int *)(cpu_string + 4)) = cpu_id[3]; + *((int *)(cpu_string + 8)) = cpu_id[2]; + __cpuid(cpu_id, 0x80000000); + u32 max_ex_fn = cpu_id[0]; + if (!strcmp(cpu_string, "GenuineIntel")) + vendor = VENDOR_INTEL; + else if (!strcmp(cpu_string, "AuthenticAMD")) + vendor = VENDOR_AMD; + else + vendor = VENDOR_OTHER; + + // Set reasonable default brand string even if brand string not available. + strcpy(brand_string, cpu_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); + 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] >> 25) & 1) bAES = true; + + // To check DAZ support, we first need to check FXSAVE support. + if ((cpu_id[3] >> 24) & 1) + { + // We can use FXSAVE. + bFXSR = true; + + GC_ALIGNED16(u8 fx_state[512]); + memset(fx_state, 0, sizeof(fx_state)); +#ifdef _WIN32 +#ifdef _M_IX86 + _fxsave(fx_state); +#elif defined (_M_X64) + _fxsave64(fx_state); +#endif +#else + __asm__("fxsave %0" : "=m" (fx_state)); +#endif + + // lowest byte of MXCSR_MASK + if ((fx_state[0x1C] >> 6) & 1) + { + // On x86, the FTZ field (supported since SSE1) only flushes denormal _outputs_ to zero, + // now that we checked DAZ support (flushing denormal _inputs_ to zero), + // we can set our generic flag. + bFlushToZero = 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_ex_fn >= 0x80000004) { + // Extract brand string + __cpuid(cpu_id, 0x80000002); + memcpy(brand_string, cpu_id, sizeof(cpu_id)); + __cpuid(cpu_id, 0x80000003); + memcpy(brand_string + 16, cpu_id, sizeof(cpu_id)); + __cpuid(cpu_id, 0x80000004); + memcpy(brand_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[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) { + __cpuid(cpu_id, 0x00000004); + 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); + 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 (bFMA) sum += ", FMA"; + if (bAES) sum += ", AES"; + if (bLongMode) sum += ", 64-bit support"; + return sum; +} + +bool CPUInfo::IsUnsafe() +{ + return !bFlushToZero; +} |
