#define CPUID_SSSE3_ECX_BIT 9
#define CPUID_OSXSAVE_ECX_BIT 27
#define CPUID_AVX_ECX_BIT 28
#define CPUID_AVX2_EBX_BIT 5
#define CPUID_SSE_EDX_BIT 25
#define BIT_XMM_STATE (1 << 1)
#define BIT_YMM_STATE (2 << 1)
#define REG_EAX 0
#define REG_EBX 1
#define REG_ECX 2
#define REG_EDX 3
#if defined(__GNUC__) && (defined(__i386__) ||defined(__x86_64))
#include <cpuid.h>
#define CPL_CPUID(level, subfunction, array) __cpuid_count(level, subfunction, array[0], array[1], array[2], array[3])
#elif defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_X64))
#include <intrin.h>
#define CPL_CPUID(level, subfunction, array) __cpuidex(array, level, subfunction)
#else
#error "not supported"
#endif
#if defined(__GNUC__) && (defined(__i386__) ||defined(__x86_64))
int CPLHaveRuntimeAVX()
{
int cpuinfo[4] = { 0, 0, 0, 0 };
unsigned int nXCRLow;
unsigned int nXCRHigh;
CPL_CPUID(1, 0, cpuinfo);
if ((cpuinfo[REG_ECX] & (1 << CPUID_OSXSAVE_ECX_BIT)) == 0) {
return 0;
}
if ((cpuinfo[REG_ECX] & (1 << CPUID_AVX_ECX_BIT)) == 0) {
return 0;
}
__asm__("xgetbv" : "=a"(nXCRLow), "=d"(nXCRHigh) : "c"(0));
if ((nXCRLow & (BIT_XMM_STATE | BIT_YMM_STATE)) !=
(BIT_XMM_STATE | BIT_YMM_STATE)) {
return 0;
}
return 1;
}
#elif defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219) && (defined(_M_IX86) || defined(_M_X64))
int CPLHaveRuntimeAVX()
{
int cpuinfo[4] = { 0, 0, 0, 0 };
unsigned __int64 xcrFeatureMask;
CPL_CPUID(1, 0, cpuinfo);
if ((cpuinfo[REG_ECX] & (1 << CPUID_OSXSAVE_ECX_BIT)) == 0) {
return 0;
}
if ((cpuinfo[REG_ECX] & (1 << CPUID_AVX_ECX_BIT)) == 0) {
return 0;
}
xcrFeatureMask = _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
if ((xcrFeatureMask & (BIT_XMM_STATE | BIT_YMM_STATE)) !=
(BIT_XMM_STATE | BIT_YMM_STATE)) {
return 0;
}
return 1;
}
#endif
int CPLHaveRuntimeAVX2()
{
int cpuinfo[4] = { 0, 0, 0, 0 };
if (!CPLHaveRuntimeAVX()) {
return 0;
}
CPL_CPUID(7, 0, cpuinfo);
if ((cpuinfo[REG_EBX] & (1 << CPUID_AVX2_EBX_BIT)) == 0) {
return 0;
}
return 1;
}
int main()
{
if (CPLHaveRuntimeAVX2()) {
return 0;
}
return 1;
}