//===================================================================
void IppLoader::initializeIpp(){
bool bVerbose = true;
IppStatus status = ippStaticInitCpu(ippGetCpuType());
if (bVerbose) cout << status << " " << ippStsNoErr << " " << ippStsCpuMismatch << " " << ippStsNoOperationInDll << endl;
/* ippStsNoErr - required target cpu library code is successfully set
// ippStsCpuMismatch - required target cpu library can't be set, the previous set is used
// ippStsNoOperationInDll
*/
const IppLibraryVersion* lib = ippiGetLibVersion();
if (bVerbose) printf("%s %s %d.%d.%d.%d\n", lib->Name, lib->Version,lib->major, lib->minor, lib->majorBuild, lib->build);
IppCpuType cpuType;
IppStatus pStatus;
Ipp64u pFeatureMask;
Ipp32u pCpuidInfoRegs[4];
cpuType=ippGetCpuType();
pStatus=ippGetCpuFeatures(&pFeatureMask, pCpuidInfoRegs);
if (bVerbose) printCpuType(cpuType);
if (bVerbose) printCpuCapability(pFeatureMask);
}
SInt32 CAVectorUnit_Examine()
{
int result = kVecNone;
#if TARGET_OS_WIN32
#if HAS_IPP
// Initialize the static IPP library! This needs to be done before
// any IPP function calls, otherwise we may have a performance penalty
int status = ippStaticInit();
if ( status == ippStsNonIntelCpu )
{
IppCpuType cpuType = ippGetCpuType();
if ( cpuType >= ippCpuSSE || cpuType <= ippCpuSSE42 )
ippStaticInitCpu( cpuType );
}
#endif
{
// On Windows we use cpuid to detect the vector unit because it works on Intel and AMD.
// The IPP library does not detect SSE on AMD processors.
if (IsCpuidAvailable())
{
if(IsSSE3Available())
{
result = kVecSSE3;
}
else if(IsSSE2Available())
{
result = kVecSSE2;
}
}
}
#elif TARGET_OS_MAC
#if DEBUG
if (getenv("CA_NoVector")) {
fprintf(stderr, "CA_NoVector set; Vector unit optimized routines will be bypassed\n");
return result;
}
else
#endif
{
#if (TARGET_CPU_PPC || TARGET_CPU_PPC64)
int sels[2] = { CTL_HW, HW_VECTORUNIT };
int vType = 0; //0 == scalar only
size_t length = sizeof(vType);
int error = sysctl(sels, 2, &vType, &length, NULL, 0);
if (!error && vType > 0)
result = kVecAltivec;
#elif (TARGET_CPU_X86 || TARGET_CPU_X86_64)
int answer = 0;
size_t length = sizeof(answer);
int error = sysctlbyname("hw.optional.sse3", &answer, &length, NULL, 0);
if (!error && answer)
result = kVecSSE3;
else {
answer = 0;
length = sizeof(answer);
error = sysctlbyname("hw.optional.sse2", &answer, &length, NULL, 0);
if (!error && answer)
result = kVecSSE2;
}
#elif (TARGET_CPU_ARM) && defined(_ARM_ARCH_7)
result = kVecNeon;
#endif
}
#endif
gCAVectorUnitType = result;
return result;
}
